Skip to content

Environments

EnvParams

Dataclass to hold environment parameters. Parameters are immutable.

Parameters:

Name Type Description Default
max_requests Scalar

Maximum number of requests in an episode

 required
incremental_loading Scalar

Incremental increase in traffic load (non-expiring requests)

 required
end_first_blocking Scalar

End episode on first blocking event

 required
continuous_operation Scalar

If True, do not reset the environment at the end of an episode

 required
edges Array

Two column array defining source-dest node-pair edges of the graph

 required
slot_size Scalar

Spectral width of frequency slot in GHz

 required
consider_modulation_format Scalar

If True, consider modulation format to determine required slots

 required
link_length_array Array

Array of link lengths

 required
aggregate_slots Scalar

Number of slots to aggregate into a single action (First-Fit with aggregation)

 required
guardband Scalar

Guard band in slots

 required
directed_graph bool

Whether graph is directed (one fibre per link per transmission direction)

 required
Source code in xlron/environments/dataclasses.py 
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
@struct.dataclass
class EnvParams:
    """Dataclass to hold environment parameters. Parameters are immutable.

    Args:
        max_requests (chex.Scalar): Maximum number of requests in an episode
        incremental_loading (chex.Scalar): Incremental increase in traffic load (non-expiring requests)
        end_first_blocking (chex.Scalar): End episode on first blocking event
        continuous_operation (chex.Scalar): If True, do not reset the environment at the end of an episode
        edges (chex.Array): Two column array defining source-dest node-pair edges of the graph
        slot_size (chex.Scalar): Spectral width of frequency slot in GHz
        consider_modulation_format (chex.Scalar): If True, consider modulation format to determine required slots
        link_length_array (chex.Array): Array of link lengths
        aggregate_slots (chex.Scalar): Number of slots to aggregate into a single action (First-Fit with aggregation)
        guardband (chex.Scalar): Guard band in slots
        directed_graph (bool): Whether graph is directed (one fibre per link per transmission direction)
    """
    max_requests: chex.Scalar = struct.field(pytree_node=False)
    incremental_loading: chex.Scalar = struct.field(pytree_node=False)
    end_first_blocking: chex.Scalar = struct.field(pytree_node=False)
    continuous_operation: chex.Scalar = struct.field(pytree_node=False)
    edges: chex.Array = struct.field(pytree_node=False)
    slot_size: chex.Scalar = struct.field(pytree_node=False)
    consider_modulation_format: chex.Scalar = struct.field(pytree_node=False)
    link_length_array: chex.Array = struct.field(pytree_node=False)
    aggregate_slots: chex.Scalar = struct.field(pytree_node=False)
    guardband: chex.Scalar = struct.field(pytree_node=False)
    directed_graph: bool = struct.field(pytree_node=False)
    maximise_throughput: bool = struct.field(pytree_node=False)
    reward_type: str = struct.field(pytree_node=False)
    values_bw: chex.Array = struct.field(pytree_node=False)
    truncate_holding_time: bool = struct.field(pytree_node=False)
    traffic_array: bool = struct.field(pytree_node=False)

EnvState

Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

Parameters:

Name Type Description Default
current_time Scalar

Current time in environment

 required
holding_time Scalar

Holding time of current request

 required
total_timesteps Scalar

Total timesteps in environment

 required
total_requests Scalar

Total requests in environment

 required
graph GraphsTuple

Graph tuple representing network state

 required
full_link_slot_mask Array

Action mask for link slot action (including if slot actions are aggregated)

 required
accepted_services Array

Number of accepted services

 required
accepted_bitrate Array

Accepted bitrate

 required
Source code in xlron/environments/dataclasses.py 
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
@struct.dataclass
class EnvState:
    """Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

    Args:
        current_time (chex.Scalar): Current time in environment
        holding_time (chex.Scalar): Holding time of current request
        total_timesteps (chex.Scalar): Total timesteps in environment
        total_requests (chex.Scalar): Total requests in environment
        graph (jraph.GraphsTuple): Graph tuple representing network state
        full_link_slot_mask (chex.Array): Action mask for link slot action (including if slot actions are aggregated)
        accepted_services (chex.Array): Number of accepted services
        accepted_bitrate (chex.Array): Accepted bitrate
        """
    current_time: chex.Scalar
    holding_time: chex.Scalar
    total_timesteps: chex.Scalar
    total_requests: chex.Scalar
    graph: jraph.GraphsTuple
    full_link_slot_mask: chex.Array
    accepted_services: chex.Array
    accepted_bitrate: chex.Array
    total_bitrate: chex.Array
    list_of_requests: chex.Array

RMSAGNModelEnv

Bases: RSAEnv

RMSA + GNN model environment.

Source code in xlron/environments/gn_model/rmsa_gn_model.py 
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
class RMSAGNModelEnv(RSAEnv):
    """RMSA + GNN model environment."""
    def __init__(
            self,
            key: chex.PRNGKey,
            params: RSAGNModelEnvParams,
            traffic_matrix: chex.Array = None,
            launch_power_array: chex.Array = None,
            list_of_requests: chex.Array = None,
            laplacian_matrix: chex.Array = None,
    ):
        """Initialise the environment state and set as initial state.

        Args:
            key: PRNG key
            params: Environment parameters
            traffic_matrix (optional): Traffic matrix
            launch_power_array (optional): Launch power array

        Returns:
            None
        """
        super().__init__(key, params, traffic_matrix=traffic_matrix, list_of_requests=list_of_requests, laplacian_matrix=laplacian_matrix)
        state = RMSAGNModelEnvState(
            current_time=0,
            holding_time=0,
            total_timesteps=0,
            total_requests=-1,
            link_slot_array=init_link_slot_array(params),
            link_slot_departure_array=init_link_slot_departure_array(params),
            request_array=init_rsa_request_array(),
            link_slot_mask=init_link_slot_mask(params, agg=params.aggregate_slots),
            traffic_matrix=traffic_matrix if traffic_matrix is not None else init_traffic_matrix(key, params),
            graph=None,
            full_link_slot_mask=init_link_slot_mask(params),
            accepted_services=0,
            accepted_bitrate=0.,
            total_bitrate=0.,
            list_of_requests=list_of_requests,
            link_snr_array=init_link_snr_array(params),
            path_index_array=init_path_index_array(params),
            path_index_array_prev=init_path_index_array(params),
            channel_centre_bw_array=init_channel_centre_bw_array(params),
            channel_power_array=init_channel_power_array(params),
            modulation_format_index_array=init_modulation_format_index_array(params),
            channel_centre_bw_array_prev=init_channel_centre_bw_array(params),
            channel_power_array_prev=init_channel_power_array(params),
            modulation_format_index_array_prev=init_modulation_format_index_array(params),
            launch_power_array=launch_power_array,
            mod_format_mask=init_mod_format_mask(params),
        )
        self.initial_state = state.replace(graph=init_graph_tuple(state, params, laplacian_matrix))

    @partial(jax.jit, static_argnums=(0, 2,))
    def action_mask(self, state: RSAEnvState, params: RSAEnvParams) -> RSAEnvState:
        """Returns mask of valid actions.

        Args:
            state: Environment state
            params: Environment parameters

        Returns:
            state: Environment state with action mask
        """
        state = mask_slots_rmsa_gn_model(state, params, state.request_array)
        return state

    @partial(jax.jit, static_argnums=(0, 2,))
    def get_obs(self, state: RMSAGNModelEnvState, params: RMSAGNModelEnvParams) -> chex.Array:
        return get_paths_obs_gn_model(state, params)

    @staticmethod
    def num_actions(params: RSAEnvParams) -> int:
        """Number of actions possible in environment."""
        return 1

    def observation_space(self, params: RSAEnvParams):
        """Observation space of the environment."""
        return spaces.Discrete(
            3 +  # Request array
            1 +  # Holding time
            7 * params.k_paths
            # Path stats:
            # Mean free block size
            # Free slots
            # Path length (100 km)
            # Path length (hops)
            # Number of connections on path
            # Mean power of connection on path
            # Mean SNR of connection on path
        )

__init__(key, params, traffic_matrix=None, launch_power_array=None, list_of_requests=None, laplacian_matrix=None)

Initialise the environment state and set as initial state.

Parameters:

Name Type Description Default
key PRNGKey

PRNG key

 required
params RSAGNModelEnvParams

Environment parameters

 required
traffic_matrix optional

Traffic matrix

 None
launch_power_array optional

Launch power array

 None

Returns:

Type Description

None
Source code in xlron/environments/gn_model/rmsa_gn_model.py 
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
def __init__(
        self,
        key: chex.PRNGKey,
        params: RSAGNModelEnvParams,
        traffic_matrix: chex.Array = None,
        launch_power_array: chex.Array = None,
        list_of_requests: chex.Array = None,
        laplacian_matrix: chex.Array = None,
):
    """Initialise the environment state and set as initial state.

    Args:
        key: PRNG key
        params: Environment parameters
        traffic_matrix (optional): Traffic matrix
        launch_power_array (optional): Launch power array

    Returns:
        None
    """
    super().__init__(key, params, traffic_matrix=traffic_matrix, list_of_requests=list_of_requests, laplacian_matrix=laplacian_matrix)
    state = RMSAGNModelEnvState(
        current_time=0,
        holding_time=0,
        total_timesteps=0,
        total_requests=-1,
        link_slot_array=init_link_slot_array(params),
        link_slot_departure_array=init_link_slot_departure_array(params),
        request_array=init_rsa_request_array(),
        link_slot_mask=init_link_slot_mask(params, agg=params.aggregate_slots),
        traffic_matrix=traffic_matrix if traffic_matrix is not None else init_traffic_matrix(key, params),
        graph=None,
        full_link_slot_mask=init_link_slot_mask(params),
        accepted_services=0,
        accepted_bitrate=0.,
        total_bitrate=0.,
        list_of_requests=list_of_requests,
        link_snr_array=init_link_snr_array(params),
        path_index_array=init_path_index_array(params),
        path_index_array_prev=init_path_index_array(params),
        channel_centre_bw_array=init_channel_centre_bw_array(params),
        channel_power_array=init_channel_power_array(params),
        modulation_format_index_array=init_modulation_format_index_array(params),
        channel_centre_bw_array_prev=init_channel_centre_bw_array(params),
        channel_power_array_prev=init_channel_power_array(params),
        modulation_format_index_array_prev=init_modulation_format_index_array(params),
        launch_power_array=launch_power_array,
        mod_format_mask=init_mod_format_mask(params),
    )
    self.initial_state = state.replace(graph=init_graph_tuple(state, params, laplacian_matrix))

action_mask(state, params)

Returns mask of valid actions.

Parameters:

Name Type Description Default
state RSAEnvState

Environment state

 required
params RSAEnvParams

Environment parameters

 required

Returns:

Name Type Description
state RSAEnvState

Environment state with action mask
Source code in xlron/environments/gn_model/rmsa_gn_model.py 
66
67
68
69
70
71
72
73
74
75
76
77
78
@partial(jax.jit, static_argnums=(0, 2,))
def action_mask(self, state: RSAEnvState, params: RSAEnvParams) -> RSAEnvState:
    """Returns mask of valid actions.

    Args:
        state: Environment state
        params: Environment parameters

    Returns:
        state: Environment state with action mask
    """
    state = mask_slots_rmsa_gn_model(state, params, state.request_array)
    return state

num_actions(params) staticmethod

Number of actions possible in environment.

Source code in xlron/environments/gn_model/rmsa_gn_model.py 
84
85
86
87
@staticmethod
def num_actions(params: RSAEnvParams) -> int:
    """Number of actions possible in environment."""
    return 1

observation_space(params)

Observation space of the environment.

Source code in xlron/environments/gn_model/rmsa_gn_model.py 
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
def observation_space(self, params: RSAEnvParams):
    """Observation space of the environment."""
    return spaces.Discrete(
        3 +  # Request array
        1 +  # Holding time
        7 * params.k_paths
        # Path stats:
        # Mean free block size
        # Free slots
        # Path length (100 km)
        # Path length (hops)
        # Number of connections on path
        # Mean power of connection on path
        # Mean SNR of connection on path
    )

RMSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
306
307
308
309
310
311
312
313
@struct.dataclass
class RMSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulations_array: chex.Array = struct.field(pytree_node=False)

RMSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
316
317
318
319
320
321
322
323
324
325
@struct.dataclass
class RMSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulation_format_index_array: chex.Array  # Modulation format index for each active connection
    modulation_format_index_array_prev: chex.Array  # Modulation format index for each active connection in previous timestep
    mod_format_mask: chex.Array  # Modulation format mask

RSAGNModelEnv

Bases: RSAEnv

RMSA + GNN model environment.

Source code in xlron/environments/gn_model/rsa_gn_model.py 
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
class RSAGNModelEnv(RSAEnv):
    """RMSA + GNN model environment."""
    def __init__(
            self,
            key: chex.PRNGKey,
            params: RSAGNModelEnvParams,
            traffic_matrix: chex.Array = None,
            launch_power_array: chex.Array = None,
            list_of_requests: chex.Array = None,
            laplacian_matrix: chex.Array = None,
    ):
        """Initialise the environment state and set as initial state.

        Args:
            key: PRNG key
            params: Environment parameters
            traffic_matrix (optional): Traffic matrix
            launch_power_array (optional): Launch power array

        Returns:
            None
        """
        super().__init__(key, params, traffic_matrix=traffic_matrix, list_of_requests=list_of_requests, laplacian_matrix=laplacian_matrix)
        state = RSAGNModelEnvState(
            current_time=0,
            holding_time=0,
            total_timesteps=0,
            total_requests=-1,
            link_slot_array=init_link_slot_array(params),
            link_slot_departure_array=init_link_slot_departure_array(params),
            request_array=init_rsa_request_array(),
            link_slot_mask=init_link_slot_mask(params, agg=params.aggregate_slots),
            traffic_matrix=traffic_matrix if traffic_matrix is not None else init_traffic_matrix(key, params),
            graph=None,
            full_link_slot_mask=init_link_slot_mask(params),
            accepted_services=0,
            accepted_bitrate=0.,
            total_bitrate=0.,
            list_of_requests=list_of_requests,
            link_snr_array=init_link_snr_array(params),
            path_index_array=init_path_index_array(params),
            path_index_array_prev=init_path_index_array(params),
            channel_centre_bw_array=init_channel_centre_bw_array(params),
            channel_power_array=init_channel_power_array(params),
            channel_centre_bw_array_prev=init_channel_centre_bw_array(params),
            channel_power_array_prev=init_channel_power_array(params),
            launch_power_array=launch_power_array,
            active_lightpaths_array=init_active_lightpaths_array(params),
            active_lightpaths_array_departure=init_active_lightpaths_array_departure(params),
            current_throughput=0.,
        )
        self.initial_state = state.replace(graph=init_graph_tuple(state, params, laplacian_matrix))

    @partial(jax.jit, static_argnums=(0, 2,))
    def get_obs(self, state: RSAGNModelEnvState, params: RSAGNModelEnvParams) -> chex.Array:
        return get_paths_obs_gn_model(state, params)

    @staticmethod
    def num_actions(params: RSAEnvParams) -> int:
        """Number of actions possible in environment."""
        return 1

    def observation_space(self, params: RSAEnvParams):
        """Observation space of the environment."""
        return spaces.Discrete(
            3 +  # Request array
            1 +  # Holding time
            7 * params.k_paths
            # Path stats:
            # Mean free block size
            # Free slots
            # Path length (100 km)
            # Path length (hops)
            # Number of connections on path
            # Mean power of connection on path
            # Mean SNR of connection on path
        )

__init__(key, params, traffic_matrix=None, launch_power_array=None, list_of_requests=None, laplacian_matrix=None)

Initialise the environment state and set as initial state.

Parameters:

Name Type Description Default
key PRNGKey

PRNG key

 required
params RSAGNModelEnvParams

Environment parameters

 required
traffic_matrix optional

Traffic matrix

 None
launch_power_array optional

Launch power array

 None

Returns:

Type Description

None
Source code in xlron/environments/gn_model/rsa_gn_model.py 
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
def __init__(
        self,
        key: chex.PRNGKey,
        params: RSAGNModelEnvParams,
        traffic_matrix: chex.Array = None,
        launch_power_array: chex.Array = None,
        list_of_requests: chex.Array = None,
        laplacian_matrix: chex.Array = None,
):
    """Initialise the environment state and set as initial state.

    Args:
        key: PRNG key
        params: Environment parameters
        traffic_matrix (optional): Traffic matrix
        launch_power_array (optional): Launch power array

    Returns:
        None
    """
    super().__init__(key, params, traffic_matrix=traffic_matrix, list_of_requests=list_of_requests, laplacian_matrix=laplacian_matrix)
    state = RSAGNModelEnvState(
        current_time=0,
        holding_time=0,
        total_timesteps=0,
        total_requests=-1,
        link_slot_array=init_link_slot_array(params),
        link_slot_departure_array=init_link_slot_departure_array(params),
        request_array=init_rsa_request_array(),
        link_slot_mask=init_link_slot_mask(params, agg=params.aggregate_slots),
        traffic_matrix=traffic_matrix if traffic_matrix is not None else init_traffic_matrix(key, params),
        graph=None,
        full_link_slot_mask=init_link_slot_mask(params),
        accepted_services=0,
        accepted_bitrate=0.,
        total_bitrate=0.,
        list_of_requests=list_of_requests,
        link_snr_array=init_link_snr_array(params),
        path_index_array=init_path_index_array(params),
        path_index_array_prev=init_path_index_array(params),
        channel_centre_bw_array=init_channel_centre_bw_array(params),
        channel_power_array=init_channel_power_array(params),
        channel_centre_bw_array_prev=init_channel_centre_bw_array(params),
        channel_power_array_prev=init_channel_power_array(params),
        launch_power_array=launch_power_array,
        active_lightpaths_array=init_active_lightpaths_array(params),
        active_lightpaths_array_departure=init_active_lightpaths_array_departure(params),
        current_throughput=0.,
    )
    self.initial_state = state.replace(graph=init_graph_tuple(state, params, laplacian_matrix))

num_actions(params) staticmethod

Number of actions possible in environment.

Source code in xlron/environments/gn_model/rsa_gn_model.py 
70
71
72
73
@staticmethod
def num_actions(params: RSAEnvParams) -> int:
    """Number of actions possible in environment."""
    return 1

observation_space(params)

Observation space of the environment.

Source code in xlron/environments/gn_model/rsa_gn_model.py 
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
def observation_space(self, params: RSAEnvParams):
    """Observation space of the environment."""
    return spaces.Discrete(
        3 +  # Request array
        1 +  # Holding time
        7 * params.k_paths
        # Path stats:
        # Mean free block size
        # Free slots
        # Path length (100 km)
        # Path length (hops)
        # Number of connections on path
        # Mean power of connection on path
        # Mean SNR of connection on path
    )

RSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
290
291
292
293
294
@struct.dataclass
class RSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RSA with GN model.
    """
    pass

RSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
297
298
299
300
301
302
303
@struct.dataclass
class RSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RSA with GN model.
    """
    active_lightpaths_array: chex.Array  # Active lightpath array. 1 x M array. Each value is a lightpath index. Used to calculate total throughput.
    active_lightpaths_array_departure: chex.Array  # Active lightpath array departure time.
    current_throughput: chex.Array  # Current throughput

Dataclasses

DeepRMSAEnvState

Bases: RSAEnvState

Dataclass to hold environment state for DeepRMSA.

Parameters:

Name Type Description Default
path_stats Array

Path stats array containing

 required
Source code in xlron/environments/dataclasses.py 
191
192
193
194
195
196
197
198
199
200
201
202
@struct.dataclass
class DeepRMSAEnvState(RSAEnvState):
    """Dataclass to hold environment state for DeepRMSA.

    Args:
        path_stats (chex.Array): Path stats array containing
        1. Required slots on path
        2. Total available slots on path
        3. Size of 1st free spectrum block
        4. Avg. free block size
    """
    path_stats: chex.Array

EnvParams

Dataclass to hold environment parameters. Parameters are immutable.

Parameters:

Name Type Description Default
max_requests Scalar

Maximum number of requests in an episode

 required
incremental_loading Scalar

Incremental increase in traffic load (non-expiring requests)

 required
end_first_blocking Scalar

End episode on first blocking event

 required
continuous_operation Scalar

If True, do not reset the environment at the end of an episode

 required
edges Array

Two column array defining source-dest node-pair edges of the graph

 required
slot_size Scalar

Spectral width of frequency slot in GHz

 required
consider_modulation_format Scalar

If True, consider modulation format to determine required slots

 required
link_length_array Array

Array of link lengths

 required
aggregate_slots Scalar

Number of slots to aggregate into a single action (First-Fit with aggregation)

 required
guardband Scalar

Guard band in slots

 required
directed_graph bool

Whether graph is directed (one fibre per link per transmission direction)

 required
Source code in xlron/environments/dataclasses.py 
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
@struct.dataclass
class EnvParams:
    """Dataclass to hold environment parameters. Parameters are immutable.

    Args:
        max_requests (chex.Scalar): Maximum number of requests in an episode
        incremental_loading (chex.Scalar): Incremental increase in traffic load (non-expiring requests)
        end_first_blocking (chex.Scalar): End episode on first blocking event
        continuous_operation (chex.Scalar): If True, do not reset the environment at the end of an episode
        edges (chex.Array): Two column array defining source-dest node-pair edges of the graph
        slot_size (chex.Scalar): Spectral width of frequency slot in GHz
        consider_modulation_format (chex.Scalar): If True, consider modulation format to determine required slots
        link_length_array (chex.Array): Array of link lengths
        aggregate_slots (chex.Scalar): Number of slots to aggregate into a single action (First-Fit with aggregation)
        guardband (chex.Scalar): Guard band in slots
        directed_graph (bool): Whether graph is directed (one fibre per link per transmission direction)
    """
    max_requests: chex.Scalar = struct.field(pytree_node=False)
    incremental_loading: chex.Scalar = struct.field(pytree_node=False)
    end_first_blocking: chex.Scalar = struct.field(pytree_node=False)
    continuous_operation: chex.Scalar = struct.field(pytree_node=False)
    edges: chex.Array = struct.field(pytree_node=False)
    slot_size: chex.Scalar = struct.field(pytree_node=False)
    consider_modulation_format: chex.Scalar = struct.field(pytree_node=False)
    link_length_array: chex.Array = struct.field(pytree_node=False)
    aggregate_slots: chex.Scalar = struct.field(pytree_node=False)
    guardband: chex.Scalar = struct.field(pytree_node=False)
    directed_graph: bool = struct.field(pytree_node=False)
    maximise_throughput: bool = struct.field(pytree_node=False)
    reward_type: str = struct.field(pytree_node=False)
    values_bw: chex.Array = struct.field(pytree_node=False)
    truncate_holding_time: bool = struct.field(pytree_node=False)
    traffic_array: bool = struct.field(pytree_node=False)

EnvState

Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

Parameters:

Name Type Description Default
current_time Scalar

Current time in environment

 required
holding_time Scalar

Holding time of current request

 required
total_timesteps Scalar

Total timesteps in environment

 required
total_requests Scalar

Total requests in environment

 required
graph GraphsTuple

Graph tuple representing network state

 required
full_link_slot_mask Array

Action mask for link slot action (including if slot actions are aggregated)

 required
accepted_services Array

Number of accepted services

 required
accepted_bitrate Array

Accepted bitrate

 required
Source code in xlron/environments/dataclasses.py 
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
@struct.dataclass
class EnvState:
    """Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

    Args:
        current_time (chex.Scalar): Current time in environment
        holding_time (chex.Scalar): Holding time of current request
        total_timesteps (chex.Scalar): Total timesteps in environment
        total_requests (chex.Scalar): Total requests in environment
        graph (jraph.GraphsTuple): Graph tuple representing network state
        full_link_slot_mask (chex.Array): Action mask for link slot action (including if slot actions are aggregated)
        accepted_services (chex.Array): Number of accepted services
        accepted_bitrate (chex.Array): Accepted bitrate
        """
    current_time: chex.Scalar
    holding_time: chex.Scalar
    total_timesteps: chex.Scalar
    total_requests: chex.Scalar
    graph: jraph.GraphsTuple
    full_link_slot_mask: chex.Array
    accepted_services: chex.Array
    accepted_bitrate: chex.Array
    total_bitrate: chex.Array
    list_of_requests: chex.Array

GNModelEnvParams

Bases: RSAEnvParams

Dataclass to hold environment state for GN model environments.

Source code in xlron/environments/dataclasses.py 
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
@struct.dataclass
class GNModelEnvParams(RSAEnvParams):
    """Dataclass to hold environment state for GN model environments.
    """
    ref_lambda: chex.Scalar = struct.field(pytree_node=False)
    max_spans: chex.Scalar = struct.field(pytree_node=False)
    max_span_length: chex.Scalar = struct.field(pytree_node=False)
    nonlinear_coeff: chex.Scalar = struct.field(pytree_node=False)
    raman_gain_slope: chex.Scalar = struct.field(pytree_node=False)
    attenuation: chex.Scalar = struct.field(pytree_node=False)
    attenuation_bar: chex.Scalar = struct.field(pytree_node=False)
    dispersion_coeff: chex.Scalar = struct.field(pytree_node=False)
    dispersion_slope: chex.Scalar = struct.field(pytree_node=False)
    noise_figure: chex.Scalar = struct.field(pytree_node=False)
    coherent: bool = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    num_roadms: chex.Scalar = struct.field(pytree_node=False)
    roadm_loss: chex.Scalar = struct.field(pytree_node=False)
    num_spans: chex.Scalar = struct.field(pytree_node=False)
    launch_power_type: chex.Scalar = struct.field(pytree_node=False)
    snr_margin: chex.Scalar = struct.field(pytree_node=False)
    max_snr: chex.Scalar = struct.field(pytree_node=False)
    max_power: chex.Scalar = struct.field(pytree_node=False)
    min_power: chex.Scalar = struct.field(pytree_node=False)
    step_power: chex.Scalar = struct.field(pytree_node=False)
    last_fit: bool = struct.field(pytree_node=False)
    default_launch_power: chex.Scalar = struct.field(pytree_node=False)
    mod_format_correction: bool = struct.field(pytree_node=False)

GNModelEnvState

Bases: RSAEnvState

Dataclass to hold environment state for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
276
277
278
279
280
281
282
283
284
285
286
287
@struct.dataclass
class GNModelEnvState(RSAEnvState):
    """Dataclass to hold environment state for RSA with GN model.
    """
    link_snr_array: chex.Array  # Available SNR on each link
    channel_centre_bw_array: chex.Array  # Channel centre bandwidth for each active connection
    path_index_array: chex.Array  # Contains indices of lightpaths in use on slots (used for lightpath SNR calculation)
    channel_power_array: chex.Array  # Channel power for each active connection
    channel_centre_bw_array_prev: chex.Array  # Channel centre bandwidth for each active connection in previous timestep
    path_index_array_prev: chex.Array  # Contains indices of lightpaths in use on slots in previous timestep
    channel_power_array_prev: chex.Array  # Channel power for each active connection in previous timestep
    launch_power_array: chex.Array  # Launch power array

LogEnvState

Dataclass to hold environment state for logging.

Parameters:

Name Type Description Default
env_state EnvState

Environment state

 required
lengths Scalar

Lengths

 required
returns Scalar

Returns

 required
cum_returns Scalar

Cumulative returns

 required
episode_lengths Scalar

Episode lengths

 required
episode_returns Scalar

Episode returns

 required
accepted_services Scalar

Accepted services

 required
accepted_bitrate Scalar

Accepted bitrate

 required
done Scalar

Done

 required
Source code in xlron/environments/dataclasses.py 
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
@struct.dataclass
class LogEnvState:
    """Dataclass to hold environment state for logging.

    Args:
        env_state (EnvState): Environment state
        lengths (chex.Scalar): Lengths
        returns (chex.Scalar): Returns
        cum_returns (chex.Scalar): Cumulative returns
        episode_lengths (chex.Scalar): Episode lengths
        episode_returns (chex.Scalar): Episode returns
        accepted_services (chex.Scalar): Accepted services
        accepted_bitrate (chex.Scalar): Accepted bitrate
        done (chex.Scalar): Done
    """
    env_state: EnvState
    lengths: float
    returns: float
    cum_returns: float
    accepted_services: int
    accepted_bitrate: float
    total_bitrate: float
    utilisation: float
    done: bool

MultiBandRSAEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
237
238
239
240
241
242
@struct.dataclass
class MultiBandRSAEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for MultiBandRSA (RBSA).
    """
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)

MultiBandRSAEnvState

Bases: RSAEnvState

Dataclass to hold environment state for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
230
231
232
233
234
@struct.dataclass
class MultiBandRSAEnvState(RSAEnvState):
    """Dataclass to hold environment state for MultiBandRSA (RBSA).
    """
    pass

RMSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
306
307
308
309
310
311
312
313
@struct.dataclass
class RMSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulations_array: chex.Array = struct.field(pytree_node=False)

RMSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
316
317
318
319
320
321
322
323
324
325
@struct.dataclass
class RMSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulation_format_index_array: chex.Array  # Modulation format index for each active connection
    modulation_format_index_array_prev: chex.Array  # Modulation format index for each active connection in previous timestep
    mod_format_mask: chex.Array  # Modulation format mask

RSAEnvParams

Bases: EnvParams

Dataclass to hold environment parameters for RSA.

Parameters:

Name Type Description Default
num_nodes Scalar

Number of nodes

 required
num_links Scalar

Number of links

 required
link_resources Scalar

Number of link resources

 required
k_paths Scalar

Number of paths

 required
mean_service_holding_time Scalar

Mean service holding time

 required
load Scalar

Load

 required
arrival_rate Scalar

Arrival rate

 required
path_link_array Array

Path link array

 required
random_traffic bool

Random traffic matrix for RSA on each reset (else uniform or custom)

 required
max_slots Scalar

Maximum number of slots

 required
path_se_array Array

Path spectral efficiency array

 required
deterministic_requests bool

If True, use deterministic requests

 required
multiple_topologies bool

If True, use multiple topologies

 required
Source code in xlron/environments/dataclasses.py 
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
@struct.dataclass
class RSAEnvParams(EnvParams):
    """Dataclass to hold environment parameters for RSA.

    Args:
        num_nodes (chex.Scalar): Number of nodes
        num_links (chex.Scalar): Number of links
        link_resources (chex.Scalar): Number of link resources
        k_paths (chex.Scalar): Number of paths
        mean_service_holding_time (chex.Scalar): Mean service holding time
        load (chex.Scalar): Load
        arrival_rate (chex.Scalar): Arrival rate
        path_link_array (chex.Array): Path link array
        random_traffic (bool): Random traffic matrix for RSA on each reset (else uniform or custom)
        max_slots (chex.Scalar): Maximum number of slots
        path_se_array (chex.Array): Path spectral efficiency array
        deterministic_requests (bool): If True, use deterministic requests
        multiple_topologies (bool): If True, use multiple topologies
    """
    num_nodes: chex.Scalar = struct.field(pytree_node=False)
    num_links: chex.Scalar = struct.field(pytree_node=False)
    link_resources: chex.Scalar = struct.field(pytree_node=False)
    k_paths: chex.Scalar = struct.field(pytree_node=False)
    mean_service_holding_time: chex.Scalar = struct.field(pytree_node=False)
    load: chex.Scalar = struct.field(pytree_node=False)
    arrival_rate: chex.Scalar = struct.field(pytree_node=False)
    path_link_array: chex.Array = struct.field(pytree_node=False)
    random_traffic: bool = struct.field(pytree_node=False)
    max_slots: chex.Scalar = struct.field(pytree_node=False)
    path_se_array: chex.Array = struct.field(pytree_node=False)
    deterministic_requests: bool = struct.field(pytree_node=False)
    multiple_topologies: bool = struct.field(pytree_node=False)
    log_actions: bool = struct.field(pytree_node=False)
    disable_node_features: bool = struct.field(pytree_node=False)

RSAEnvState

Bases: EnvState

Dataclass to hold environment state for RSA.

Parameters:

Name Type Description Default
link_slot_array Array

Link slot array

 required
request_array Array

Request array

 required
link_slot_departure_array Array

Link slot departure array

 required
link_slot_mask Array

Link slot mask

 required
traffic_matrix Array

Traffic matrix

 required
Source code in xlron/environments/dataclasses.py 
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
@struct.dataclass
class RSAEnvState(EnvState):
    """Dataclass to hold environment state for RSA.

    Args:
        link_slot_array (chex.Array): Link slot array
        request_array (chex.Array): Request array
        link_slot_departure_array (chex.Array): Link slot departure array
        link_slot_mask (chex.Array): Link slot mask
        traffic_matrix (chex.Array): Traffic matrix
    """
    link_slot_array: chex.Array
    request_array: chex.Array
    link_slot_departure_array: chex.Array
    link_slot_mask: chex.Array
    traffic_matrix: chex.Array

RSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
290
291
292
293
294
@struct.dataclass
class RSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RSA with GN model.
    """
    pass

RSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
297
298
299
300
301
302
303
@struct.dataclass
class RSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RSA with GN model.
    """
    active_lightpaths_array: chex.Array  # Active lightpath array. 1 x M array. Each value is a lightpath index. Used to calculate total throughput.
    active_lightpaths_array_departure: chex.Array  # Active lightpath array departure time.
    current_throughput: chex.Array  # Current throughput

RSAMultibandEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
335
336
337
338
339
340
@struct.dataclass
class RSAMultibandEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for MultiBandRSA (RBSA).
    """
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)

RSAMultibandEnvState

Bases: RSAEnvState

Dataclass to hold environment state for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
328
329
330
331
332
@struct.dataclass
class RSAMultibandEnvState(RSAEnvState):
    """Dataclass to hold environment state for MultiBandRSA (RBSA).
    """
    pass

RWALightpathReuseEnvState

Bases: RSAEnvState

Dataclass to hold environment state for RWA with lightpath reuse.

Parameters:

Name Type Description Default
path_index_array Array

Contains indices of lightpaths in use on slots

 required
path_capacity_array Array

Contains remaining capacity of each lightpath

 required
link_capacity_array Array

Contains remaining capacity of lightpath on each link-slot

 required
Source code in xlron/environments/dataclasses.py 
210
211
212
213
214
215
216
217
218
219
220
221
222
@struct.dataclass
class RWALightpathReuseEnvState(RSAEnvState):
    """Dataclass to hold environment state for RWA with lightpath reuse.

    Args:
        path_index_array (chex.Array): Contains indices of lightpaths in use on slots
        path_capacity_array (chex.Array): Contains remaining capacity of each lightpath
        link_capacity_array (chex.Array): Contains remaining capacity of lightpath on each link-slot
    """
    path_index_array: chex.Array  # Contains indices of lightpaths in use on slots
    path_capacity_array: chex.Array  # Contains remaining capacity of each lightpath
    link_capacity_array: chex.Array  # Contains remaining capacity of lightpath on each link-slot
    time_since_last_departure: chex.Array  # Time since last departure

VONEEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for VONE.

Parameters:

Name Type Description Default
node_resources Scalar

Number of node resources

 required
max_edges Scalar

Maximum number of edges

 required
min_node_resources Scalar

Minimum number of node resources

 required
max_node_resources Scalar

Maximum number of node resources

 required
Source code in xlron/environments/dataclasses.py 
369
370
371
372
373
374
375
376
377
378
379
380
381
382
@struct.dataclass
class VONEEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for VONE.

    Args:
        node_resources (chex.Scalar): Number of node resources
        max_edges (chex.Scalar): Maximum number of edges
        min_node_resources (chex.Scalar): Minimum number of node resources
        max_node_resources (chex.Scalar): Maximum number of node resources
    """
    node_resources: chex.Scalar = struct.field(pytree_node=False)
    max_edges: chex.Scalar = struct.field(pytree_node=False)
    min_node_resources: chex.Scalar = struct.field(pytree_node=False)
    max_node_resources: chex.Scalar = struct.field(pytree_node=False)

VONEEnvState

Bases: RSAEnvState

Dataclass to hold environment state for VONE.

Parameters:

Name Type Description Default
node_capacity_array Array

Node capacity array

 required
node_resource_array Array

Node resource array

 required
node_departure_array Array

Node departure array

 required
action_counter Array

Action counter

 required
action_history Array

Action history

 required
node_mask_s Array

Node mask for source node

 required
node_mask_d Array

Node mask for destination node

 required
virtual_topology_patterns Array

Virtual topology patterns

 required
values_nodes Array

Values for nodes

 required
Source code in xlron/environments/dataclasses.py 
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
@struct.dataclass
class VONEEnvState(RSAEnvState):
    """Dataclass to hold environment state for VONE.

    Args:
        node_capacity_array (chex.Array): Node capacity array
        node_resource_array (chex.Array): Node resource array
        node_departure_array (chex.Array): Node departure array
        action_counter (chex.Array): Action counter
        action_history (chex.Array): Action history
        node_mask_s (chex.Array): Node mask for source node
        node_mask_d (chex.Array): Node mask for destination node
        virtual_topology_patterns (chex.Array): Virtual topology patterns
        values_nodes (chex.Array): Values for nodes
    """
    node_capacity_array: chex.Array
    node_resource_array: chex.Array
    node_departure_array: chex.Array
    action_counter: chex.Array
    action_history: chex.Array
    node_mask_s: chex.Array
    node_mask_d: chex.Array
    virtual_topology_patterns: chex.Array
    values_nodes: chex.Array

Environment wrappers

DeepRMSAEnvState

Bases: RSAEnvState

Dataclass to hold environment state for DeepRMSA.

Parameters:

Name Type Description Default
path_stats Array

Path stats array containing

 required
Source code in xlron/environments/dataclasses.py 
191
192
193
194
195
196
197
198
199
200
201
202
@struct.dataclass
class DeepRMSAEnvState(RSAEnvState):
    """Dataclass to hold environment state for DeepRMSA.

    Args:
        path_stats (chex.Array): Path stats array containing
        1. Required slots on path
        2. Total available slots on path
        3. Size of 1st free spectrum block
        4. Avg. free block size
    """
    path_stats: chex.Array

EnvParams

Dataclass to hold environment parameters. Parameters are immutable.

Parameters:

Name Type Description Default
max_requests Scalar

Maximum number of requests in an episode

 required
incremental_loading Scalar

Incremental increase in traffic load (non-expiring requests)

 required
end_first_blocking Scalar

End episode on first blocking event

 required
continuous_operation Scalar

If True, do not reset the environment at the end of an episode

 required
edges Array

Two column array defining source-dest node-pair edges of the graph

 required
slot_size Scalar

Spectral width of frequency slot in GHz

 required
consider_modulation_format Scalar

If True, consider modulation format to determine required slots

 required
link_length_array Array

Array of link lengths

 required
aggregate_slots Scalar

Number of slots to aggregate into a single action (First-Fit with aggregation)

 required
guardband Scalar

Guard band in slots

 required
directed_graph bool

Whether graph is directed (one fibre per link per transmission direction)

 required
Source code in xlron/environments/dataclasses.py 
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
@struct.dataclass
class EnvParams:
    """Dataclass to hold environment parameters. Parameters are immutable.

    Args:
        max_requests (chex.Scalar): Maximum number of requests in an episode
        incremental_loading (chex.Scalar): Incremental increase in traffic load (non-expiring requests)
        end_first_blocking (chex.Scalar): End episode on first blocking event
        continuous_operation (chex.Scalar): If True, do not reset the environment at the end of an episode
        edges (chex.Array): Two column array defining source-dest node-pair edges of the graph
        slot_size (chex.Scalar): Spectral width of frequency slot in GHz
        consider_modulation_format (chex.Scalar): If True, consider modulation format to determine required slots
        link_length_array (chex.Array): Array of link lengths
        aggregate_slots (chex.Scalar): Number of slots to aggregate into a single action (First-Fit with aggregation)
        guardband (chex.Scalar): Guard band in slots
        directed_graph (bool): Whether graph is directed (one fibre per link per transmission direction)
    """
    max_requests: chex.Scalar = struct.field(pytree_node=False)
    incremental_loading: chex.Scalar = struct.field(pytree_node=False)
    end_first_blocking: chex.Scalar = struct.field(pytree_node=False)
    continuous_operation: chex.Scalar = struct.field(pytree_node=False)
    edges: chex.Array = struct.field(pytree_node=False)
    slot_size: chex.Scalar = struct.field(pytree_node=False)
    consider_modulation_format: chex.Scalar = struct.field(pytree_node=False)
    link_length_array: chex.Array = struct.field(pytree_node=False)
    aggregate_slots: chex.Scalar = struct.field(pytree_node=False)
    guardband: chex.Scalar = struct.field(pytree_node=False)
    directed_graph: bool = struct.field(pytree_node=False)
    maximise_throughput: bool = struct.field(pytree_node=False)
    reward_type: str = struct.field(pytree_node=False)
    values_bw: chex.Array = struct.field(pytree_node=False)
    truncate_holding_time: bool = struct.field(pytree_node=False)
    traffic_array: bool = struct.field(pytree_node=False)

EnvState

Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

Parameters:

Name Type Description Default
current_time Scalar

Current time in environment

 required
holding_time Scalar

Holding time of current request

 required
total_timesteps Scalar

Total timesteps in environment

 required
total_requests Scalar

Total requests in environment

 required
graph GraphsTuple

Graph tuple representing network state

 required
full_link_slot_mask Array

Action mask for link slot action (including if slot actions are aggregated)

 required
accepted_services Array

Number of accepted services

 required
accepted_bitrate Array

Accepted bitrate

 required
Source code in xlron/environments/dataclasses.py 
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
@struct.dataclass
class EnvState:
    """Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

    Args:
        current_time (chex.Scalar): Current time in environment
        holding_time (chex.Scalar): Holding time of current request
        total_timesteps (chex.Scalar): Total timesteps in environment
        total_requests (chex.Scalar): Total requests in environment
        graph (jraph.GraphsTuple): Graph tuple representing network state
        full_link_slot_mask (chex.Array): Action mask for link slot action (including if slot actions are aggregated)
        accepted_services (chex.Array): Number of accepted services
        accepted_bitrate (chex.Array): Accepted bitrate
        """
    current_time: chex.Scalar
    holding_time: chex.Scalar
    total_timesteps: chex.Scalar
    total_requests: chex.Scalar
    graph: jraph.GraphsTuple
    full_link_slot_mask: chex.Array
    accepted_services: chex.Array
    accepted_bitrate: chex.Array
    total_bitrate: chex.Array
    list_of_requests: chex.Array

GNModelEnvParams

Bases: RSAEnvParams

Dataclass to hold environment state for GN model environments.

Source code in xlron/environments/dataclasses.py 
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
@struct.dataclass
class GNModelEnvParams(RSAEnvParams):
    """Dataclass to hold environment state for GN model environments.
    """
    ref_lambda: chex.Scalar = struct.field(pytree_node=False)
    max_spans: chex.Scalar = struct.field(pytree_node=False)
    max_span_length: chex.Scalar = struct.field(pytree_node=False)
    nonlinear_coeff: chex.Scalar = struct.field(pytree_node=False)
    raman_gain_slope: chex.Scalar = struct.field(pytree_node=False)
    attenuation: chex.Scalar = struct.field(pytree_node=False)
    attenuation_bar: chex.Scalar = struct.field(pytree_node=False)
    dispersion_coeff: chex.Scalar = struct.field(pytree_node=False)
    dispersion_slope: chex.Scalar = struct.field(pytree_node=False)
    noise_figure: chex.Scalar = struct.field(pytree_node=False)
    coherent: bool = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    num_roadms: chex.Scalar = struct.field(pytree_node=False)
    roadm_loss: chex.Scalar = struct.field(pytree_node=False)
    num_spans: chex.Scalar = struct.field(pytree_node=False)
    launch_power_type: chex.Scalar = struct.field(pytree_node=False)
    snr_margin: chex.Scalar = struct.field(pytree_node=False)
    max_snr: chex.Scalar = struct.field(pytree_node=False)
    max_power: chex.Scalar = struct.field(pytree_node=False)
    min_power: chex.Scalar = struct.field(pytree_node=False)
    step_power: chex.Scalar = struct.field(pytree_node=False)
    last_fit: bool = struct.field(pytree_node=False)
    default_launch_power: chex.Scalar = struct.field(pytree_node=False)
    mod_format_correction: bool = struct.field(pytree_node=False)

GNModelEnvState

Bases: RSAEnvState

Dataclass to hold environment state for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
276
277
278
279
280
281
282
283
284
285
286
287
@struct.dataclass
class GNModelEnvState(RSAEnvState):
    """Dataclass to hold environment state for RSA with GN model.
    """
    link_snr_array: chex.Array  # Available SNR on each link
    channel_centre_bw_array: chex.Array  # Channel centre bandwidth for each active connection
    path_index_array: chex.Array  # Contains indices of lightpaths in use on slots (used for lightpath SNR calculation)
    channel_power_array: chex.Array  # Channel power for each active connection
    channel_centre_bw_array_prev: chex.Array  # Channel centre bandwidth for each active connection in previous timestep
    path_index_array_prev: chex.Array  # Contains indices of lightpaths in use on slots in previous timestep
    channel_power_array_prev: chex.Array  # Channel power for each active connection in previous timestep
    launch_power_array: chex.Array  # Launch power array

HashableArrayWrapper

Bases: Generic[T]

Wrapper for making arrays hashable. In order to access pre-computed data, such as shortest paths between node-pairs or the constituent links of a path, within a jitted function, we need to make the arrays containing this data hashable. By defining this wrapper, we can define a hash method that returns a hash of the array's bytes, thus making the array hashable. From: https://github.com/google/jax/issues/4572#issuecomment-709677518

Source code in xlron/environments/wrappers.py 
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
class HashableArrayWrapper(Generic[T]):
    """Wrapper for making arrays hashable.
    In order to access pre-computed data, such as shortest paths between node-pairs or the constituent links of a path,
    within a jitted function, we need to make the arrays containing this data hashable. By defining this wrapper, we can
    define a __hash__ method that returns a hash of the array's bytes, thus making the array hashable.
    From: https://github.com/google/jax/issues/4572#issuecomment-709677518
    """
    def __init__(self, val: T):
        self.val = val

    def __getattribute__(self, prop):
        if prop == 'val' or prop == "__hash__" or prop == "__eq__":
            return super(HashableArrayWrapper, self).__getattribute__(prop)
        return getattr(self.val, prop)

    def __getitem__(self, key):
        return self.val[key]

    def __setitem__(self, key, val):
        self.val[key] = val

    def __hash__(self):
        return hash(self.val.tobytes())

    def __eq__(self, other):
        if isinstance(other, HashableArrayWrapper):
            return self.__hash__() == other.__hash__()

        f = getattr(self.val, "__eq__")
        return f(self, other)

LogEnvState

Dataclass to hold environment state for logging.

Parameters:

Name Type Description Default
env_state EnvState

Environment state

 required
lengths Scalar

Lengths

 required
returns Scalar

Returns

 required
cum_returns Scalar

Cumulative returns

 required
episode_lengths Scalar

Episode lengths

 required
episode_returns Scalar

Episode returns

 required
accepted_services Scalar

Accepted services

 required
accepted_bitrate Scalar

Accepted bitrate

 required
done Scalar

Done

 required
Source code in xlron/environments/dataclasses.py 
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
@struct.dataclass
class LogEnvState:
    """Dataclass to hold environment state for logging.

    Args:
        env_state (EnvState): Environment state
        lengths (chex.Scalar): Lengths
        returns (chex.Scalar): Returns
        cum_returns (chex.Scalar): Cumulative returns
        episode_lengths (chex.Scalar): Episode lengths
        episode_returns (chex.Scalar): Episode returns
        accepted_services (chex.Scalar): Accepted services
        accepted_bitrate (chex.Scalar): Accepted bitrate
        done (chex.Scalar): Done
    """
    env_state: EnvState
    lengths: float
    returns: float
    cum_returns: float
    accepted_services: int
    accepted_bitrate: float
    total_bitrate: float
    utilisation: float
    done: bool

LogWrapper

Bases: GymnaxWrapper

Log the episode returns and lengths. Modified from: https://github.com/RobertTLange/gymnax/blob/master/gymnax/wrappers/purerl.py

Source code in xlron/environments/wrappers.py 
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
class LogWrapper(GymnaxWrapper):
    """Log the episode returns and lengths.
    Modified from: https://github.com/RobertTLange/gymnax/blob/master/gymnax/wrappers/purerl.py
    """

    def __init__(self, env: environment.Environment):
        super().__init__(env)

    @partial(jax.jit, static_argnums=(0,))
    def reset(
        self, key: chex.PRNGKey, params: Optional[environment.EnvParams] = None
    ) -> Tuple[chex.Array, environment.EnvState]:
        obs, env_state = self._env.reset(key, params)
        state = LogEnvState(env_state, 0, 0, 0, 0, 0, 0, 0, False)
        return obs, state

    @partial(jax.jit, static_argnums=(0,))
    def step(
        self,
        key: chex.PRNGKey,
        state: environment.EnvState,
        action: Union[int, float],
        params: Optional[environment.EnvParams] = None,
    ) -> Tuple[chex.Array, environment.EnvState, float, bool, dict]:
        obs, env_state, reward, done, info = self._env.step(
            key, state.env_state, action, params
        )
        state = LogEnvState(
            env_state=env_state,
            lengths=state.lengths * (1 - done) + 1,
            returns=reward,
            cum_returns=state.cum_returns * (1 - done) + reward,
            accepted_services=env_state.accepted_services,
            accepted_bitrate=env_state.accepted_bitrate,
            total_bitrate=env_state.total_bitrate,
            utilisation=jnp.count_nonzero(env_state.link_slot_array) / env_state.link_slot_array.size,
            done=done,
        )
        info["lengths"] = state.lengths
        info["returns"] = state.returns
        info["cum_returns"] = state.cum_returns
        info["accepted_services"] = state.accepted_services
        info["accepted_bitrate"] = state.accepted_bitrate
        info["total_bitrate"] = state.total_bitrate
        info["utilisation"] = state.utilisation
        if params.__class__.__name__ == "RSAGNModelEnvParams":
            action, power_action = action
            nodes_sd, dr_request = read_rsa_request(state.env_state.request_array)
            source, dest = nodes_sd
            i = get_path_indices(source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph).astype(
                jnp.int32)
            path_index, slot_index = process_path_action(state.env_state, params, action)
            info["launch_power"] = power_action
            info["path_index"] = i + path_index
            info["slot_index"] = slot_index
            info["source"] = source
            info["dest"] = dest
            info["data_rate"] = dr_request[0]
            if params.log_actions:
                path = params.path_link_array.val[path_index.astype(jnp.int32)]
                info["path_snr"] = get_snr_for_path(path, env_state.link_snr_array, params)[slot_index.astype(jnp.int32)]
        # Log path length if required, to calculate average path length and number of hops
        if params.log_actions:
            nodes_sd, dr_request = read_rsa_request(state.env_state.request_array)
            source, dest = nodes_sd
            i = get_path_indices(source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph).astype(
                jnp.int32)
            path_index, slot_index = process_path_action(state.env_state, params, action)
            info["path_index"] = i + path_index
            info["slot_index"] = slot_index
            info["source"] = source
            info["dest"] = dest
            info["data_rate"] = dr_request[0]
            info["arrival_time"] = env_state.current_time[0]
            info["departure_time"] = env_state.current_time[0] + env_state.holding_time[0]
        info["done"] = done
        return obs, state, reward, done, info

    def _tree_flatten(self):
        children = ()  # arrays / dynamic values
        aux_data = {"env": self._env}  # static values
        return (children, aux_data)

    @classmethod
    def _tree_unflatten(cls, aux_data, children):
        return cls(*children, **aux_data)

MultiBandRSAEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
237
238
239
240
241
242
@struct.dataclass
class MultiBandRSAEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for MultiBandRSA (RBSA).
    """
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)

MultiBandRSAEnvState

Bases: RSAEnvState

Dataclass to hold environment state for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
230
231
232
233
234
@struct.dataclass
class MultiBandRSAEnvState(RSAEnvState):
    """Dataclass to hold environment state for MultiBandRSA (RBSA).
    """
    pass

RMSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
306
307
308
309
310
311
312
313
@struct.dataclass
class RMSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulations_array: chex.Array = struct.field(pytree_node=False)

RMSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
316
317
318
319
320
321
322
323
324
325
@struct.dataclass
class RMSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulation_format_index_array: chex.Array  # Modulation format index for each active connection
    modulation_format_index_array_prev: chex.Array  # Modulation format index for each active connection in previous timestep
    mod_format_mask: chex.Array  # Modulation format mask

RSAEnvParams

Bases: EnvParams

Dataclass to hold environment parameters for RSA.

Parameters:

Name Type Description Default
num_nodes Scalar

Number of nodes

 required
num_links Scalar

Number of links

 required
link_resources Scalar

Number of link resources

 required
k_paths Scalar

Number of paths

 required
mean_service_holding_time Scalar

Mean service holding time

 required
load Scalar

Load

 required
arrival_rate Scalar

Arrival rate

 required
path_link_array Array

Path link array

 required
random_traffic bool

Random traffic matrix for RSA on each reset (else uniform or custom)

 required
max_slots Scalar

Maximum number of slots

 required
path_se_array Array

Path spectral efficiency array

 required
deterministic_requests bool

If True, use deterministic requests

 required
multiple_topologies bool

If True, use multiple topologies

 required
Source code in xlron/environments/dataclasses.py 
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
@struct.dataclass
class RSAEnvParams(EnvParams):
    """Dataclass to hold environment parameters for RSA.

    Args:
        num_nodes (chex.Scalar): Number of nodes
        num_links (chex.Scalar): Number of links
        link_resources (chex.Scalar): Number of link resources
        k_paths (chex.Scalar): Number of paths
        mean_service_holding_time (chex.Scalar): Mean service holding time
        load (chex.Scalar): Load
        arrival_rate (chex.Scalar): Arrival rate
        path_link_array (chex.Array): Path link array
        random_traffic (bool): Random traffic matrix for RSA on each reset (else uniform or custom)
        max_slots (chex.Scalar): Maximum number of slots
        path_se_array (chex.Array): Path spectral efficiency array
        deterministic_requests (bool): If True, use deterministic requests
        multiple_topologies (bool): If True, use multiple topologies
    """
    num_nodes: chex.Scalar = struct.field(pytree_node=False)
    num_links: chex.Scalar = struct.field(pytree_node=False)
    link_resources: chex.Scalar = struct.field(pytree_node=False)
    k_paths: chex.Scalar = struct.field(pytree_node=False)
    mean_service_holding_time: chex.Scalar = struct.field(pytree_node=False)
    load: chex.Scalar = struct.field(pytree_node=False)
    arrival_rate: chex.Scalar = struct.field(pytree_node=False)
    path_link_array: chex.Array = struct.field(pytree_node=False)
    random_traffic: bool = struct.field(pytree_node=False)
    max_slots: chex.Scalar = struct.field(pytree_node=False)
    path_se_array: chex.Array = struct.field(pytree_node=False)
    deterministic_requests: bool = struct.field(pytree_node=False)
    multiple_topologies: bool = struct.field(pytree_node=False)
    log_actions: bool = struct.field(pytree_node=False)
    disable_node_features: bool = struct.field(pytree_node=False)

RSAEnvState

Bases: EnvState

Dataclass to hold environment state for RSA.

Parameters:

Name Type Description Default
link_slot_array Array

Link slot array

 required
request_array Array

Request array

 required
link_slot_departure_array Array

Link slot departure array

 required
link_slot_mask Array

Link slot mask

 required
traffic_matrix Array

Traffic matrix

 required
Source code in xlron/environments/dataclasses.py 
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
@struct.dataclass
class RSAEnvState(EnvState):
    """Dataclass to hold environment state for RSA.

    Args:
        link_slot_array (chex.Array): Link slot array
        request_array (chex.Array): Request array
        link_slot_departure_array (chex.Array): Link slot departure array
        link_slot_mask (chex.Array): Link slot mask
        traffic_matrix (chex.Array): Traffic matrix
    """
    link_slot_array: chex.Array
    request_array: chex.Array
    link_slot_departure_array: chex.Array
    link_slot_mask: chex.Array
    traffic_matrix: chex.Array

RSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
290
291
292
293
294
@struct.dataclass
class RSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RSA with GN model.
    """
    pass

RSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
297
298
299
300
301
302
303
@struct.dataclass
class RSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RSA with GN model.
    """
    active_lightpaths_array: chex.Array  # Active lightpath array. 1 x M array. Each value is a lightpath index. Used to calculate total throughput.
    active_lightpaths_array_departure: chex.Array  # Active lightpath array departure time.
    current_throughput: chex.Array  # Current throughput

RSAMultibandEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
335
336
337
338
339
340
@struct.dataclass
class RSAMultibandEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for MultiBandRSA (RBSA).
    """
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)

RSAMultibandEnvState

Bases: RSAEnvState

Dataclass to hold environment state for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
328
329
330
331
332
@struct.dataclass
class RSAMultibandEnvState(RSAEnvState):
    """Dataclass to hold environment state for MultiBandRSA (RBSA).
    """
    pass

RWALightpathReuseEnvState

Bases: RSAEnvState

Dataclass to hold environment state for RWA with lightpath reuse.

Parameters:

Name Type Description Default
path_index_array Array

Contains indices of lightpaths in use on slots

 required
path_capacity_array Array

Contains remaining capacity of each lightpath

 required
link_capacity_array Array

Contains remaining capacity of lightpath on each link-slot

 required
Source code in xlron/environments/dataclasses.py 
210
211
212
213
214
215
216
217
218
219
220
221
222
@struct.dataclass
class RWALightpathReuseEnvState(RSAEnvState):
    """Dataclass to hold environment state for RWA with lightpath reuse.

    Args:
        path_index_array (chex.Array): Contains indices of lightpaths in use on slots
        path_capacity_array (chex.Array): Contains remaining capacity of each lightpath
        link_capacity_array (chex.Array): Contains remaining capacity of lightpath on each link-slot
    """
    path_index_array: chex.Array  # Contains indices of lightpaths in use on slots
    path_capacity_array: chex.Array  # Contains remaining capacity of each lightpath
    link_capacity_array: chex.Array  # Contains remaining capacity of lightpath on each link-slot
    time_since_last_departure: chex.Array  # Time since last departure

RolloutWrapper

Wrapper to define batch evaluation for generation parameters. Used for genetic algorithm. From: https://github.com/RobertTLange/gymnax/

Source code in xlron/environments/wrappers.py 
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
class RolloutWrapper:
    """Wrapper to define batch evaluation for generation parameters. Used for genetic algorithm.
    From: https://github.com/RobertTLange/gymnax/
    """
    def __init__(
        self,
        model_forward=None,
        env: environment.Environment = None,
        num_env_steps: Optional[int] = None,
        env_params: EnvParams = None,
    ):
        """Wrapper to define batch evaluation for generation parameters."""
        self.env = env
        # Define the RL environment & network forward function
        self.env_params = env_params
        self.model_forward = model_forward

        if num_env_steps is None:
            self.num_env_steps = self.env_params.max_requests
        else:
            self.num_env_steps = num_env_steps

    @partial(jax.jit, static_argnums=(0, 2))
    def population_rollout(self, rng_eval, policy_params):
        """Reshape parameter vector and evaluate the generation."""
        # Evaluate population of nets on gymnax task - vmap over rng & params
        pop_rollout = jax.vmap(self.batch_rollout, in_axes=(None, 0))
        return pop_rollout(rng_eval, policy_params)

    @partial(jax.jit, static_argnums=(0, 2))
    def batch_rollout(self, rng_eval, policy_params):
        """Evaluate a generation of networks on RL/Supervised/etc. task."""
        # vmap over different MC fitness evaluations for single network
        batch_rollout = jax.vmap(self.single_rollout, in_axes=(0, None))
        return batch_rollout(rng_eval, policy_params)

    @partial(jax.jit, static_argnums=(0, 2))
    def single_rollout(self, rng_input, policy_params):
        """Rollout a pendulum episode with lax.scan."""
        # Reset the environment
        rng_reset, rng_episode = jax.random.split(rng_input)
        obs, state = self.env.reset(rng_reset, self.env_params)

        def policy_step(state_input, tmp):
            """lax.scan compatible step transition in jax env."""
            obs, state, policy_params, rng, cum_reward, valid_mask = state_input
            rng, rng_step, rng_net = jax.random.split(rng, 3)
            if self.model_forward is not None:
                action = self.model_forward(policy_params, obs, rng_net)
            else:
                action = self.env.action_space(self.env_params).sample(rng_net)
            next_obs, next_state, reward, done, _ = self.env.step(
                rng_step, state, action, self.env_params
            )
            new_cum_reward = cum_reward + reward * valid_mask
            new_valid_mask = valid_mask * (1 - done)
            carry = [
                next_obs,
                next_state,
                policy_params,
                rng,
                new_cum_reward,
                new_valid_mask,
            ]
            y = [obs, action, reward, next_obs, done]
            return carry, y

        # Scan over episode step loop
        carry_out, scan_out = jax.lax.scan(
            policy_step,
            [
                obs,
                state,
                policy_params,
                rng_episode,
                jnp.array([0.0]),
                jnp.array([1.0]),
            ],
            (),
            self.num_env_steps,
        )
        # Return the sum of rewards accumulated by agent in episode rollout
        obs, action, reward, next_obs, done = scan_out
        cum_return = carry_out[-2]
        return obs, action, reward, next_obs, done, cum_return

    @property
    def input_shape(self):
        """Get the shape of the observation."""
        rng = jax.random.PRNGKey(0)
        obs, state = self.env.reset(rng, self.env_params)
        return obs.shape

input_shape property

Get the shape of the observation.

__init__(model_forward=None, env=None, num_env_steps=None, env_params=None)

Wrapper to define batch evaluation for generation parameters.

Source code in xlron/environments/wrappers.py 
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
def __init__(
    self,
    model_forward=None,
    env: environment.Environment = None,
    num_env_steps: Optional[int] = None,
    env_params: EnvParams = None,
):
    """Wrapper to define batch evaluation for generation parameters."""
    self.env = env
    # Define the RL environment & network forward function
    self.env_params = env_params
    self.model_forward = model_forward

    if num_env_steps is None:
        self.num_env_steps = self.env_params.max_requests
    else:
        self.num_env_steps = num_env_steps

batch_rollout(rng_eval, policy_params)

Evaluate a generation of networks on RL/Supervised/etc. task.

Source code in xlron/environments/wrappers.py 
183
184
185
186
187
188
@partial(jax.jit, static_argnums=(0, 2))
def batch_rollout(self, rng_eval, policy_params):
    """Evaluate a generation of networks on RL/Supervised/etc. task."""
    # vmap over different MC fitness evaluations for single network
    batch_rollout = jax.vmap(self.single_rollout, in_axes=(0, None))
    return batch_rollout(rng_eval, policy_params)

population_rollout(rng_eval, policy_params)

Reshape parameter vector and evaluate the generation.

Source code in xlron/environments/wrappers.py 
176
177
178
179
180
181
@partial(jax.jit, static_argnums=(0, 2))
def population_rollout(self, rng_eval, policy_params):
    """Reshape parameter vector and evaluate the generation."""
    # Evaluate population of nets on gymnax task - vmap over rng & params
    pop_rollout = jax.vmap(self.batch_rollout, in_axes=(None, 0))
    return pop_rollout(rng_eval, policy_params)

single_rollout(rng_input, policy_params)

Rollout a pendulum episode with lax.scan.

Source code in xlron/environments/wrappers.py 
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
@partial(jax.jit, static_argnums=(0, 2))
def single_rollout(self, rng_input, policy_params):
    """Rollout a pendulum episode with lax.scan."""
    # Reset the environment
    rng_reset, rng_episode = jax.random.split(rng_input)
    obs, state = self.env.reset(rng_reset, self.env_params)

    def policy_step(state_input, tmp):
        """lax.scan compatible step transition in jax env."""
        obs, state, policy_params, rng, cum_reward, valid_mask = state_input
        rng, rng_step, rng_net = jax.random.split(rng, 3)
        if self.model_forward is not None:
            action = self.model_forward(policy_params, obs, rng_net)
        else:
            action = self.env.action_space(self.env_params).sample(rng_net)
        next_obs, next_state, reward, done, _ = self.env.step(
            rng_step, state, action, self.env_params
        )
        new_cum_reward = cum_reward + reward * valid_mask
        new_valid_mask = valid_mask * (1 - done)
        carry = [
            next_obs,
            next_state,
            policy_params,
            rng,
            new_cum_reward,
            new_valid_mask,
        ]
        y = [obs, action, reward, next_obs, done]
        return carry, y

    # Scan over episode step loop
    carry_out, scan_out = jax.lax.scan(
        policy_step,
        [
            obs,
            state,
            policy_params,
            rng_episode,
            jnp.array([0.0]),
            jnp.array([1.0]),
        ],
        (),
        self.num_env_steps,
    )
    # Return the sum of rewards accumulated by agent in episode rollout
    obs, action, reward, next_obs, done = scan_out
    cum_return = carry_out[-2]
    return obs, action, reward, next_obs, done, cum_return

TimeIt

Context manager for timing execution of code blocks.

Source code in xlron/environments/wrappers.py 
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
class TimeIt:
    """Context manager for timing execution of code blocks."""

    def __init__(self, tag, frames=None):
        self.tag = tag
        self.frames = frames

    def __enter__(self):
        self.start = timeit.default_timer()
        return self

    def __exit__(self, *args):
        self.elapsed_secs = timeit.default_timer() - self.start
        msg = self.tag + (': Elapsed time=%.2fs' % self.elapsed_secs)
        if self.frames:
            msg += ', FPS=%.2e' % (self.frames / self.elapsed_secs)
        print(msg)

VONEEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for VONE.

Parameters:

Name Type Description Default
node_resources Scalar

Number of node resources

 required
max_edges Scalar

Maximum number of edges

 required
min_node_resources Scalar

Minimum number of node resources

 required
max_node_resources Scalar

Maximum number of node resources

 required
Source code in xlron/environments/dataclasses.py 
369
370
371
372
373
374
375
376
377
378
379
380
381
382
@struct.dataclass
class VONEEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for VONE.

    Args:
        node_resources (chex.Scalar): Number of node resources
        max_edges (chex.Scalar): Maximum number of edges
        min_node_resources (chex.Scalar): Minimum number of node resources
        max_node_resources (chex.Scalar): Maximum number of node resources
    """
    node_resources: chex.Scalar = struct.field(pytree_node=False)
    max_edges: chex.Scalar = struct.field(pytree_node=False)
    min_node_resources: chex.Scalar = struct.field(pytree_node=False)
    max_node_resources: chex.Scalar = struct.field(pytree_node=False)

VONEEnvState

Bases: RSAEnvState

Dataclass to hold environment state for VONE.

Parameters:

Name Type Description Default
node_capacity_array Array

Node capacity array

 required
node_resource_array Array

Node resource array

 required
node_departure_array Array

Node departure array

 required
action_counter Array

Action counter

 required
action_history Array

Action history

 required
node_mask_s Array

Node mask for source node

 required
node_mask_d Array

Node mask for destination node

 required
virtual_topology_patterns Array

Virtual topology patterns

 required
values_nodes Array

Values for nodes

 required
Source code in xlron/environments/dataclasses.py 
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
@struct.dataclass
class VONEEnvState(RSAEnvState):
    """Dataclass to hold environment state for VONE.

    Args:
        node_capacity_array (chex.Array): Node capacity array
        node_resource_array (chex.Array): Node resource array
        node_departure_array (chex.Array): Node departure array
        action_counter (chex.Array): Action counter
        action_history (chex.Array): Action history
        node_mask_s (chex.Array): Node mask for source node
        node_mask_d (chex.Array): Node mask for destination node
        virtual_topology_patterns (chex.Array): Virtual topology patterns
        values_nodes (chex.Array): Values for nodes
    """
    node_capacity_array: chex.Array
    node_resource_array: chex.Array
    node_departure_array: chex.Array
    action_counter: chex.Array
    action_history: chex.Array
    node_mask_s: chex.Array
    node_mask_d: chex.Array
    virtual_topology_patterns: chex.Array
    values_nodes: chex.Array

Constituent functions of environments

DeepRMSAEnvState

Bases: RSAEnvState

Dataclass to hold environment state for DeepRMSA.

Parameters:

Name Type Description Default
path_stats Array

Path stats array containing

 required
Source code in xlron/environments/dataclasses.py 
191
192
193
194
195
196
197
198
199
200
201
202
@struct.dataclass
class DeepRMSAEnvState(RSAEnvState):
    """Dataclass to hold environment state for DeepRMSA.

    Args:
        path_stats (chex.Array): Path stats array containing
        1. Required slots on path
        2. Total available slots on path
        3. Size of 1st free spectrum block
        4. Avg. free block size
    """
    path_stats: chex.Array

EnvParams

Dataclass to hold environment parameters. Parameters are immutable.

Parameters:

Name Type Description Default
max_requests Scalar

Maximum number of requests in an episode

 required
incremental_loading Scalar

Incremental increase in traffic load (non-expiring requests)

 required
end_first_blocking Scalar

End episode on first blocking event

 required
continuous_operation Scalar

If True, do not reset the environment at the end of an episode

 required
edges Array

Two column array defining source-dest node-pair edges of the graph

 required
slot_size Scalar

Spectral width of frequency slot in GHz

 required
consider_modulation_format Scalar

If True, consider modulation format to determine required slots

 required
link_length_array Array

Array of link lengths

 required
aggregate_slots Scalar

Number of slots to aggregate into a single action (First-Fit with aggregation)

 required
guardband Scalar

Guard band in slots

 required
directed_graph bool

Whether graph is directed (one fibre per link per transmission direction)

 required
Source code in xlron/environments/dataclasses.py 
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
@struct.dataclass
class EnvParams:
    """Dataclass to hold environment parameters. Parameters are immutable.

    Args:
        max_requests (chex.Scalar): Maximum number of requests in an episode
        incremental_loading (chex.Scalar): Incremental increase in traffic load (non-expiring requests)
        end_first_blocking (chex.Scalar): End episode on first blocking event
        continuous_operation (chex.Scalar): If True, do not reset the environment at the end of an episode
        edges (chex.Array): Two column array defining source-dest node-pair edges of the graph
        slot_size (chex.Scalar): Spectral width of frequency slot in GHz
        consider_modulation_format (chex.Scalar): If True, consider modulation format to determine required slots
        link_length_array (chex.Array): Array of link lengths
        aggregate_slots (chex.Scalar): Number of slots to aggregate into a single action (First-Fit with aggregation)
        guardband (chex.Scalar): Guard band in slots
        directed_graph (bool): Whether graph is directed (one fibre per link per transmission direction)
    """
    max_requests: chex.Scalar = struct.field(pytree_node=False)
    incremental_loading: chex.Scalar = struct.field(pytree_node=False)
    end_first_blocking: chex.Scalar = struct.field(pytree_node=False)
    continuous_operation: chex.Scalar = struct.field(pytree_node=False)
    edges: chex.Array = struct.field(pytree_node=False)
    slot_size: chex.Scalar = struct.field(pytree_node=False)
    consider_modulation_format: chex.Scalar = struct.field(pytree_node=False)
    link_length_array: chex.Array = struct.field(pytree_node=False)
    aggregate_slots: chex.Scalar = struct.field(pytree_node=False)
    guardband: chex.Scalar = struct.field(pytree_node=False)
    directed_graph: bool = struct.field(pytree_node=False)
    maximise_throughput: bool = struct.field(pytree_node=False)
    reward_type: str = struct.field(pytree_node=False)
    values_bw: chex.Array = struct.field(pytree_node=False)
    truncate_holding_time: bool = struct.field(pytree_node=False)
    traffic_array: bool = struct.field(pytree_node=False)

EnvState

Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

Parameters:

Name Type Description Default
current_time Scalar

Current time in environment

 required
holding_time Scalar

Holding time of current request

 required
total_timesteps Scalar

Total timesteps in environment

 required
total_requests Scalar

Total requests in environment

 required
graph GraphsTuple

Graph tuple representing network state

 required
full_link_slot_mask Array

Action mask for link slot action (including if slot actions are aggregated)

 required
accepted_services Array

Number of accepted services

 required
accepted_bitrate Array

Accepted bitrate

 required
Source code in xlron/environments/dataclasses.py 
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
@struct.dataclass
class EnvState:
    """Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

    Args:
        current_time (chex.Scalar): Current time in environment
        holding_time (chex.Scalar): Holding time of current request
        total_timesteps (chex.Scalar): Total timesteps in environment
        total_requests (chex.Scalar): Total requests in environment
        graph (jraph.GraphsTuple): Graph tuple representing network state
        full_link_slot_mask (chex.Array): Action mask for link slot action (including if slot actions are aggregated)
        accepted_services (chex.Array): Number of accepted services
        accepted_bitrate (chex.Array): Accepted bitrate
        """
    current_time: chex.Scalar
    holding_time: chex.Scalar
    total_timesteps: chex.Scalar
    total_requests: chex.Scalar
    graph: jraph.GraphsTuple
    full_link_slot_mask: chex.Array
    accepted_services: chex.Array
    accepted_bitrate: chex.Array
    total_bitrate: chex.Array
    list_of_requests: chex.Array

GNModelEnvParams

Bases: RSAEnvParams

Dataclass to hold environment state for GN model environments.

Source code in xlron/environments/dataclasses.py 
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
@struct.dataclass
class GNModelEnvParams(RSAEnvParams):
    """Dataclass to hold environment state for GN model environments.
    """
    ref_lambda: chex.Scalar = struct.field(pytree_node=False)
    max_spans: chex.Scalar = struct.field(pytree_node=False)
    max_span_length: chex.Scalar = struct.field(pytree_node=False)
    nonlinear_coeff: chex.Scalar = struct.field(pytree_node=False)
    raman_gain_slope: chex.Scalar = struct.field(pytree_node=False)
    attenuation: chex.Scalar = struct.field(pytree_node=False)
    attenuation_bar: chex.Scalar = struct.field(pytree_node=False)
    dispersion_coeff: chex.Scalar = struct.field(pytree_node=False)
    dispersion_slope: chex.Scalar = struct.field(pytree_node=False)
    noise_figure: chex.Scalar = struct.field(pytree_node=False)
    coherent: bool = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    num_roadms: chex.Scalar = struct.field(pytree_node=False)
    roadm_loss: chex.Scalar = struct.field(pytree_node=False)
    num_spans: chex.Scalar = struct.field(pytree_node=False)
    launch_power_type: chex.Scalar = struct.field(pytree_node=False)
    snr_margin: chex.Scalar = struct.field(pytree_node=False)
    max_snr: chex.Scalar = struct.field(pytree_node=False)
    max_power: chex.Scalar = struct.field(pytree_node=False)
    min_power: chex.Scalar = struct.field(pytree_node=False)
    step_power: chex.Scalar = struct.field(pytree_node=False)
    last_fit: bool = struct.field(pytree_node=False)
    default_launch_power: chex.Scalar = struct.field(pytree_node=False)
    mod_format_correction: bool = struct.field(pytree_node=False)

GNModelEnvState

Bases: RSAEnvState

Dataclass to hold environment state for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
276
277
278
279
280
281
282
283
284
285
286
287
@struct.dataclass
class GNModelEnvState(RSAEnvState):
    """Dataclass to hold environment state for RSA with GN model.
    """
    link_snr_array: chex.Array  # Available SNR on each link
    channel_centre_bw_array: chex.Array  # Channel centre bandwidth for each active connection
    path_index_array: chex.Array  # Contains indices of lightpaths in use on slots (used for lightpath SNR calculation)
    channel_power_array: chex.Array  # Channel power for each active connection
    channel_centre_bw_array_prev: chex.Array  # Channel centre bandwidth for each active connection in previous timestep
    path_index_array_prev: chex.Array  # Contains indices of lightpaths in use on slots in previous timestep
    channel_power_array_prev: chex.Array  # Channel power for each active connection in previous timestep
    launch_power_array: chex.Array  # Launch power array

LogEnvState

Dataclass to hold environment state for logging.

Parameters:

Name Type Description Default
env_state EnvState

Environment state

 required
lengths Scalar

Lengths

 required
returns Scalar

Returns

 required
cum_returns Scalar

Cumulative returns

 required
episode_lengths Scalar

Episode lengths

 required
episode_returns Scalar

Episode returns

 required
accepted_services Scalar

Accepted services

 required
accepted_bitrate Scalar

Accepted bitrate

 required
done Scalar

Done

 required
Source code in xlron/environments/dataclasses.py 
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
@struct.dataclass
class LogEnvState:
    """Dataclass to hold environment state for logging.

    Args:
        env_state (EnvState): Environment state
        lengths (chex.Scalar): Lengths
        returns (chex.Scalar): Returns
        cum_returns (chex.Scalar): Cumulative returns
        episode_lengths (chex.Scalar): Episode lengths
        episode_returns (chex.Scalar): Episode returns
        accepted_services (chex.Scalar): Accepted services
        accepted_bitrate (chex.Scalar): Accepted bitrate
        done (chex.Scalar): Done
    """
    env_state: EnvState
    lengths: float
    returns: float
    cum_returns: float
    accepted_services: int
    accepted_bitrate: float
    total_bitrate: float
    utilisation: float
    done: bool

MultiBandRSAEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
237
238
239
240
241
242
@struct.dataclass
class MultiBandRSAEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for MultiBandRSA (RBSA).
    """
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)

MultiBandRSAEnvState

Bases: RSAEnvState

Dataclass to hold environment state for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
230
231
232
233
234
@struct.dataclass
class MultiBandRSAEnvState(RSAEnvState):
    """Dataclass to hold environment state for MultiBandRSA (RBSA).
    """
    pass

RMSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
306
307
308
309
310
311
312
313
@struct.dataclass
class RMSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulations_array: chex.Array = struct.field(pytree_node=False)

RMSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
316
317
318
319
320
321
322
323
324
325
@struct.dataclass
class RMSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulation_format_index_array: chex.Array  # Modulation format index for each active connection
    modulation_format_index_array_prev: chex.Array  # Modulation format index for each active connection in previous timestep
    mod_format_mask: chex.Array  # Modulation format mask

RSAEnvParams

Bases: EnvParams

Dataclass to hold environment parameters for RSA.

Parameters:

Name Type Description Default
num_nodes Scalar

Number of nodes

 required
num_links Scalar

Number of links

 required
link_resources Scalar

Number of link resources

 required
k_paths Scalar

Number of paths

 required
mean_service_holding_time Scalar

Mean service holding time

 required
load Scalar

Load

 required
arrival_rate Scalar

Arrival rate

 required
path_link_array Array

Path link array

 required
random_traffic bool

Random traffic matrix for RSA on each reset (else uniform or custom)

 required
max_slots Scalar

Maximum number of slots

 required
path_se_array Array

Path spectral efficiency array

 required
deterministic_requests bool

If True, use deterministic requests

 required
multiple_topologies bool

If True, use multiple topologies

 required
Source code in xlron/environments/dataclasses.py 
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
@struct.dataclass
class RSAEnvParams(EnvParams):
    """Dataclass to hold environment parameters for RSA.

    Args:
        num_nodes (chex.Scalar): Number of nodes
        num_links (chex.Scalar): Number of links
        link_resources (chex.Scalar): Number of link resources
        k_paths (chex.Scalar): Number of paths
        mean_service_holding_time (chex.Scalar): Mean service holding time
        load (chex.Scalar): Load
        arrival_rate (chex.Scalar): Arrival rate
        path_link_array (chex.Array): Path link array
        random_traffic (bool): Random traffic matrix for RSA on each reset (else uniform or custom)
        max_slots (chex.Scalar): Maximum number of slots
        path_se_array (chex.Array): Path spectral efficiency array
        deterministic_requests (bool): If True, use deterministic requests
        multiple_topologies (bool): If True, use multiple topologies
    """
    num_nodes: chex.Scalar = struct.field(pytree_node=False)
    num_links: chex.Scalar = struct.field(pytree_node=False)
    link_resources: chex.Scalar = struct.field(pytree_node=False)
    k_paths: chex.Scalar = struct.field(pytree_node=False)
    mean_service_holding_time: chex.Scalar = struct.field(pytree_node=False)
    load: chex.Scalar = struct.field(pytree_node=False)
    arrival_rate: chex.Scalar = struct.field(pytree_node=False)
    path_link_array: chex.Array = struct.field(pytree_node=False)
    random_traffic: bool = struct.field(pytree_node=False)
    max_slots: chex.Scalar = struct.field(pytree_node=False)
    path_se_array: chex.Array = struct.field(pytree_node=False)
    deterministic_requests: bool = struct.field(pytree_node=False)
    multiple_topologies: bool = struct.field(pytree_node=False)
    log_actions: bool = struct.field(pytree_node=False)
    disable_node_features: bool = struct.field(pytree_node=False)

RSAEnvState

Bases: EnvState

Dataclass to hold environment state for RSA.

Parameters:

Name Type Description Default
link_slot_array Array

Link slot array

 required
request_array Array

Request array

 required
link_slot_departure_array Array

Link slot departure array

 required
link_slot_mask Array

Link slot mask

 required
traffic_matrix Array

Traffic matrix

 required
Source code in xlron/environments/dataclasses.py 
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
@struct.dataclass
class RSAEnvState(EnvState):
    """Dataclass to hold environment state for RSA.

    Args:
        link_slot_array (chex.Array): Link slot array
        request_array (chex.Array): Request array
        link_slot_departure_array (chex.Array): Link slot departure array
        link_slot_mask (chex.Array): Link slot mask
        traffic_matrix (chex.Array): Traffic matrix
    """
    link_slot_array: chex.Array
    request_array: chex.Array
    link_slot_departure_array: chex.Array
    link_slot_mask: chex.Array
    traffic_matrix: chex.Array

RSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
290
291
292
293
294
@struct.dataclass
class RSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RSA with GN model.
    """
    pass

RSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
297
298
299
300
301
302
303
@struct.dataclass
class RSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RSA with GN model.
    """
    active_lightpaths_array: chex.Array  # Active lightpath array. 1 x M array. Each value is a lightpath index. Used to calculate total throughput.
    active_lightpaths_array_departure: chex.Array  # Active lightpath array departure time.
    current_throughput: chex.Array  # Current throughput

RSAMultibandEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
335
336
337
338
339
340
@struct.dataclass
class RSAMultibandEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for MultiBandRSA (RBSA).
    """
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)

RSAMultibandEnvState

Bases: RSAEnvState

Dataclass to hold environment state for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
328
329
330
331
332
@struct.dataclass
class RSAMultibandEnvState(RSAEnvState):
    """Dataclass to hold environment state for MultiBandRSA (RBSA).
    """
    pass

RWALightpathReuseEnvState

Bases: RSAEnvState

Dataclass to hold environment state for RWA with lightpath reuse.

Parameters:

Name Type Description Default
path_index_array Array

Contains indices of lightpaths in use on slots

 required
path_capacity_array Array

Contains remaining capacity of each lightpath

 required
link_capacity_array Array

Contains remaining capacity of lightpath on each link-slot

 required
Source code in xlron/environments/dataclasses.py 
210
211
212
213
214
215
216
217
218
219
220
221
222
@struct.dataclass
class RWALightpathReuseEnvState(RSAEnvState):
    """Dataclass to hold environment state for RWA with lightpath reuse.

    Args:
        path_index_array (chex.Array): Contains indices of lightpaths in use on slots
        path_capacity_array (chex.Array): Contains remaining capacity of each lightpath
        link_capacity_array (chex.Array): Contains remaining capacity of lightpath on each link-slot
    """
    path_index_array: chex.Array  # Contains indices of lightpaths in use on slots
    path_capacity_array: chex.Array  # Contains remaining capacity of each lightpath
    link_capacity_array: chex.Array  # Contains remaining capacity of lightpath on each link-slot
    time_since_last_departure: chex.Array  # Time since last departure

VONEEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for VONE.

Parameters:

Name Type Description Default
node_resources Scalar

Number of node resources

 required
max_edges Scalar

Maximum number of edges

 required
min_node_resources Scalar

Minimum number of node resources

 required
max_node_resources Scalar

Maximum number of node resources

 required
Source code in xlron/environments/dataclasses.py 
369
370
371
372
373
374
375
376
377
378
379
380
381
382
@struct.dataclass
class VONEEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for VONE.

    Args:
        node_resources (chex.Scalar): Number of node resources
        max_edges (chex.Scalar): Maximum number of edges
        min_node_resources (chex.Scalar): Minimum number of node resources
        max_node_resources (chex.Scalar): Maximum number of node resources
    """
    node_resources: chex.Scalar = struct.field(pytree_node=False)
    max_edges: chex.Scalar = struct.field(pytree_node=False)
    min_node_resources: chex.Scalar = struct.field(pytree_node=False)
    max_node_resources: chex.Scalar = struct.field(pytree_node=False)

VONEEnvState

Bases: RSAEnvState

Dataclass to hold environment state for VONE.

Parameters:

Name Type Description Default
node_capacity_array Array

Node capacity array

 required
node_resource_array Array

Node resource array

 required
node_departure_array Array

Node departure array

 required
action_counter Array

Action counter

 required
action_history Array

Action history

 required
node_mask_s Array

Node mask for source node

 required
node_mask_d Array

Node mask for destination node

 required
virtual_topology_patterns Array

Virtual topology patterns

 required
values_nodes Array

Values for nodes

 required
Source code in xlron/environments/dataclasses.py 
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
@struct.dataclass
class VONEEnvState(RSAEnvState):
    """Dataclass to hold environment state for VONE.

    Args:
        node_capacity_array (chex.Array): Node capacity array
        node_resource_array (chex.Array): Node resource array
        node_departure_array (chex.Array): Node departure array
        action_counter (chex.Array): Action counter
        action_history (chex.Array): Action history
        node_mask_s (chex.Array): Node mask for source node
        node_mask_d (chex.Array): Node mask for destination node
        virtual_topology_patterns (chex.Array): Virtual topology patterns
        values_nodes (chex.Array): Values for nodes
    """
    node_capacity_array: chex.Array
    node_resource_array: chex.Array
    node_departure_array: chex.Array
    action_counter: chex.Array
    action_history: chex.Array
    node_mask_s: chex.Array
    node_mask_d: chex.Array
    virtual_topology_patterns: chex.Array
    values_nodes: chex.Array

aggregate_slots(full_mask, params)

Aggregate slot mask to reduce action space. Only used if the --aggregate_slots flag is set to > 1. Aggregated action is valid if there is one valid slot action within the aggregated action window.

Parameters:

Name Type Description Default
full_mask Array

slot mask

 required
params EnvParams

environment parameters

 required

Returns:

Name Type Description
agg_mask Array

aggregated slot mask
Source code in xlron/environments/env_funcs.py 
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
@partial(jax.jit, static_argnums=(1,))
def aggregate_slots(full_mask: chex.Array, params: EnvParams) -> chex.Array:
    """Aggregate slot mask to reduce action space. Only used if the --aggregate_slots flag is set to > 1.
    Aggregated action is valid if there is one valid slot action within the aggregated action window.

    Args:
        full_mask: slot mask
        params: environment parameters

    Returns:
        agg_mask: aggregated slot mask
    """

    num_actions = math.ceil(params.link_resources/params.aggregate_slots)
    agg_mask = jnp.zeros((params.k_paths, num_actions), dtype=jnp.float32)

    def get_max(i, mask_val):
        """Get maximum value of array slice of length aggregate_slots."""
        mask_slice = jax.lax.dynamic_slice(
                mask_val,
                (0, i * params.aggregate_slots,),
                (1,  params.aggregate_slots,),
            )
        max_slice = jnp.max(mask_slice).reshape(1, -1)
        return max_slice

    def update_window_max(i, val):
        """Update ith index 'agg_mask' with max of ith slice of length aggregate_slots from 'full_mask'.

        Args:
            i: increments as += aggregate_slots
            val: tuple of (agg_mask, path_mask, path_index).
        Returns:
            new_agg_mask: agg_mask is updated with max of path_mask for window size aggregate_slots
            mask: mask is unchanged
            path_index: path_index is unchanged
        """
        agg_mask = val[0]
        full_mask = val[1]
        path_index = val[2]
        new_agg_mask = jax.lax.dynamic_update_slice(
            agg_mask,
            get_max(i, full_mask),
            (path_index, i),
        )
        return new_agg_mask, full_mask, path_index

    def apply_to_path_mask(i, val):
        """
        Loop through each path for num_actions steps and get_window_max at each step.

        Args:
            i: path index
            val: tuple of (agg_mask, mask) where mask is original link-slot mask and agg_mask is resulting aggregated mask
        Returns:
            new_agg_mask: agg_mask is updated with aggregated path mask
            mask: mask is unchanged
        """
        val = (
            val[0],  # aggregated mask (to be updated)
            val[1][i].reshape(1, -1),  # mask for path i
            i  # path index
        )
        new_agg_mask = jax.lax.fori_loop(
            0,
            num_actions,
            update_window_max,
            val,
        )[0]
        return new_agg_mask, full_mask

    return jax.lax.fori_loop(
            0,
            params.k_paths,
            apply_to_path_mask,
            (agg_mask, full_mask),
        )

calculate_path_capacity(path_length, min_request=100, scale_factor=1.0, alpha=0.0002, NF=4.5, B=10000000000000.0, R_s=100000000000.0, beta_2=-2.17e-26, gamma=0.0012, L_s=100000.0, lambda0=1.55e-06)

From Nevin JOCN paper 2022: https://discovery.ucl.ac.uk/id/eprint/10175456/1/RL_JOCN_accepted.pdf

Source code in xlron/environments/env_funcs.py 
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
def calculate_path_capacity(
        path_length,
        min_request=100,  # Minimum data rate request size
        scale_factor=1.0,  # Scale factor for link capacity
        alpha=0.2e-3, # Fibre attenuation coefficient
        NF=4.5,  # Amplifier noise figure
        B=10e12,  # Total modulated bandwidth
        R_s=100e9,  # Symbol rate
        beta_2=-21.7e-27,  # Dispersion parameter
        gamma=1.2e-3,  # Nonlinear coefficient
        L_s=100e3,  # Span length
        lambda0=1550e-9,  # Wavelength
):
    """From Nevin JOCN paper 2022: https://discovery.ucl.ac.uk/id/eprint/10175456/1/RL_JOCN_accepted.pdf"""
    alpha_lin = alpha / 4.343  # Linear attenuation coefficient
    N_spans = jnp.floor(path_length * 1e3 / L_s)  # Number of fibre spans on path
    L_eff = (1 - jnp.exp(-alpha_lin * L_s)) / alpha_lin  # Effective length of span in m
    sigma_2_ase = (jnp.exp(alpha_lin * L_s) - 1) * 10**(NF/10) * 6.626e-34 * 2.998e8 * R_s / lambda0  # ASE noise power
    span_NSR = jnp.cbrt(2 * sigma_2_ase**2 * alpha_lin * gamma**2 * L_eff**2 *
                        jnp.log(jnp.pi**2 * jnp.abs(beta_2) * B**2 / alpha_lin) / (jnp.pi * jnp.abs(beta_2) * R_s**2))  # Noise-to-signal ratio per span
    path_NSR = jnp.where(N_spans < 1, 1, N_spans) * span_NSR  # Noise-to-signal ratio per path
    path_capacity = 2 * R_s/1e9 * jnp.log2(1 + 1/path_NSR)  # Capacity of path in Gbps
    # Round link capacity down to nearest increment of minimum request size and apply scale factor
    path_capacity = jnp.floor(path_capacity * scale_factor / min_request) * min_request
    return path_capacity

calculate_path_stats(state, params, request)

For use in DeepRMSA agent observation space. Calculate: 1. Size of 1st suitable free spectrum block 2. Index of 1st suitable free spectrum block 3. Required slots on path 4. Avg. free block size 5. Free slots

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required
request Array

Request array in format [source_node, data-rate, destination_node]

 required

Returns:

Name Type Description
stats Array

Array of calculated path statistics
Source code in xlron/environments/env_funcs.py 
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
@partial(jax.jit, static_argnums=(1,))
def calculate_path_stats(state: EnvState, params: EnvParams, request: chex.Array) -> chex.Array:
    """For use in DeepRMSA agent observation space.
    Calculate:
    1. Size of 1st suitable free spectrum block
    2. Index of 1st suitable free spectrum block
    3. Required slots on path
    4. Avg. free block size
    5. Free slots

    Args:
        state: Environment state
        params: Environment parameters
        request: Request array in format [source_node, data-rate, destination_node]

    Returns:
        stats: Array of calculated path statistics
    """
    nodes_sd, requested_datarate = read_rsa_request(request)
    init_val = jnp.zeros((params.k_paths, 5))
    # TODO - check if the normalisation is useful
    def body_fun(i, val):
        slots = get_path_slots(state.link_slot_array, params, nodes_sd, i)
        se = get_paths_se(params, nodes_sd)[i] if params.consider_modulation_format else jnp.array([1])
        req_slots = jnp.squeeze(required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband))
        req_slots_norm = req_slots#*params.slot_size / jnp.max(params.values_bw.val)
        free_slots_norm = jnp.sum(jnp.where(slots == 0, 1, 0)) #/ params.link_resources
        block_sizes = find_block_sizes(slots)
        first_block_index = jnp.argmax(block_sizes >= req_slots)
        first_block_index_norm = first_block_index #/ params.link_resources
        first_block_size_norm = jnp.squeeze(
            jax.lax.dynamic_slice(block_sizes, (first_block_index,), (1,))
        ) / req_slots
        avg_block_size_norm = jnp.sum(block_sizes) / jnp.max(jnp.array([jnp.sum(find_block_starts(slots)), 1])) #/ req_slots
        val = jax.lax.dynamic_update_slice(
            val,
            jnp.array([[first_block_size_norm, first_block_index_norm, req_slots_norm, avg_block_size_norm, free_slots_norm]]),
            (i, 0)
        )  # N.B. that all values are normalised
        return val

    stats = jax.lax.fori_loop(
            0,
            params.k_paths,
            body_fun,
            init_val,
        )

    return stats

check_action_rmsa_gn_model(state, action, params)

Check if action is valid for RSA GN model Args: state (EnvState): Environment state params (EnvParams): Environment parameters action (chex.Array): Action array Returns: bool: True if action is invalid, False if action is valid

Source code in xlron/environments/env_funcs.py 
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
@partial(jax.jit, static_argnums=(1,))
def check_action_rmsa_gn_model(state: EnvState, action: Optional[chex.Array], params: EnvParams) -> bool:
    """Check if action is valid for RSA GN model
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
        action (chex.Array): Action array
    Returns:
        bool: True if action is invalid, False if action is valid
    """
    # Check if action is valid
    # TODO - log failure reasons in info
    snr_sufficient_check = check_snr_sufficient(state, params)
    spectrum_reuse_check = check_no_spectrum_reuse(state.link_slot_array)
    # jax.debug.print("spectrum_reuse_check {}", spectrum_reuse_check, ordered=True)
    # jax.debug.print("snr_sufficient_check {}", snr_sufficient_check, ordered=True)
    return jnp.any(jnp.stack((
        spectrum_reuse_check,
        snr_sufficient_check,
    )))

check_action_rsa(state)

Check if action is valid. Combines checks for: - no spectrum reuse

Parameters:

Name Type Description Default
state

current state

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
def check_action_rsa(state):
    """Check if action is valid.
    Combines checks for:
    - no spectrum reuse

    Args:
        state: current state

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.any(jnp.stack((
        check_no_spectrum_reuse(state.link_slot_array),
    )))

check_action_rwalr(state, action, params)

Combines checks for: - no spectrum reuse - lightpath available and existing

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Name Type Description
bool bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
def check_action_rwalr(state: EnvState, action: chex.Array, params: EnvParams) -> bool:
    """Combines checks for:
    - no spectrum reuse
    - lightpath available and existing

    Args:
        state: Environment state

    Returns:
        bool: True if check failed, False if check passed

    """
    return jnp.any(jnp.stack((
        check_no_spectrum_reuse(state.link_slot_array),
        jnp.logical_not(check_lightpath_available_and_existing(state, params, action)[0]),
    )))

check_all_nodes_assigned(node_departure_array, total_requested_nodes)

Count negative values on each node (row) in node departure array, sum them, must equal total requested_nodes.

Parameters:

Name Type Description Default
node_departure_array Array

Node departure array (N x R) where N is number of nodes and R is number of resources

 required
total_requested_nodes int

Total requested nodes (int)

 required

Returns:

Name Type Description
bool bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
def check_all_nodes_assigned(node_departure_array: chex.Array, total_requested_nodes: int) -> bool:
    """Count negative values on each node (row) in node departure array, sum them, must equal total requested_nodes.

    Args:
        node_departure_array: Node departure array (N x R) where N is number of nodes and R is number of resources
        total_requested_nodes: Total requested nodes (int)

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.sum(jnp.sum(jnp.where(node_departure_array < 0, 1, 0), axis=1)) != total_requested_nodes

check_lightpath_available_and_existing(state, params, action)

Check if lightpath is available and existing. Available means that the lightpath does not use slots occupied by a different lightpath. Existing means that the lightpath has already been established.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required

Returns:

Name Type Description
lightpath_available_check Tuple[Array, Array, Array, Array]

True if lightpath is available
Source code in xlron/environments/env_funcs.py 
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
@partial(jax.jit, static_argnums=(1,))
def check_lightpath_available_and_existing(state: EnvState, params: EnvParams, action: chex.Array) -> (
        Tuple)[chex.Array, chex.Array, chex.Array, chex.Array]:
    """Check if lightpath is available and existing.
    Available means that the lightpath does not use slots occupied by a different lightpath.
    Existing means that the lightpath has already been established.

    Args:
        state: Environment state
        params: Environment parameters

    Returns:
        lightpath_available_check: True if lightpath is available
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_index, initial_slot_index = process_path_action(state, params, action)
    path = get_paths(params, nodes_sd)[path_index]
    # Get unique lightpath index
    lightpath_index = get_lightpath_index(params, nodes_sd, path_index)
    # Get mask for slots that lightpath will occupy
    # negative numbers used so as not to conflict with lightpath indices
    new_lightpath_mask = vmap_set_path_links(
        jnp.full((params.num_links, 1), -2), path, 0, 1, -1
    )
    path_index_array = state.path_index_array[:, initial_slot_index].reshape(-1, 1)
    masked_path_index_array = jnp.where(
        new_lightpath_mask == -1, path_index_array, -2
    )
    lightpath_mask = jnp.where(
        path_index_array == lightpath_index, -1, -2
    )  # Allow current lightpath
    lightpath_existing_check = jnp.array_equal(lightpath_mask, new_lightpath_mask)  # True if all slots are same
    lightpath_mask = jnp.where(masked_path_index_array == -1, -1, lightpath_mask)  # Allow empty slots
    # True if all slots are same or empty
    lightpath_available_check = jnp.logical_or(
        jnp.array_equal(lightpath_mask, new_lightpath_mask), lightpath_existing_check
    )
    curr_lightpath_capacity = jnp.max(
        jnp.where(new_lightpath_mask == -1, state.link_capacity_array[:, initial_slot_index].reshape(-1, 1), 0)
    )
    return lightpath_available_check, lightpath_existing_check, curr_lightpath_capacity, lightpath_index

check_min_two_nodes_assigned(node_departure_array)

Count negative values on each node (row) in node departure array, sum them, must be 2 or greater. This check is important if e.g. an action contains 2 nodes the same therefore only assigns 1 node. Return False if check passed, True if check failed

Parameters:

Name Type Description Default
node_departure_array Array

Node departure array (N x R) where N is number of nodes and R is number of resources

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
def check_min_two_nodes_assigned(node_departure_array: chex.Array):
    """Count negative values on each node (row) in node departure array, sum them, must be 2 or greater.
    This check is important if e.g. an action contains 2 nodes the same therefore only assigns 1 node.
    Return False if check passed, True if check failed

    Args:
        node_departure_array: Node departure array (N x R) where N is number of nodes and R is number of resources

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.sum(jnp.sum(jnp.where(node_departure_array < 0, 1, 0), axis=1)) <= 1

check_no_spectrum_reuse(link_slot_array)

slot-=1 when used, should be zero when unoccupied, so check if any < -1 in slot array.

Parameters:

Name Type Description Default
link_slot_array

Link slot array (L x S) where L is number of links and S is number of slots

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
def check_no_spectrum_reuse(link_slot_array):
    """slot-=1 when used, should be zero when unoccupied, so check if any < -1 in slot array.

    Args:
        link_slot_array: Link slot array (L x S) where L is number of links and S is number of slots

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.any(link_slot_array < -1)

check_node_capacities(capacity_array)

Sum selected nodes array and check less than node resources.

Parameters:

Name Type Description Default
capacity_array Array

Node capacity array (N x 1) where N is number of nodes

 required

Returns:

Name Type Description
bool bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
def check_node_capacities(capacity_array: chex.Array) -> bool:
    """Sum selected nodes array and check less than node resources.

    Args:
        capacity_array: Node capacity array (N x 1) where N is number of nodes

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.any(capacity_array < 0)

check_snr_sufficient(state, params)

Check if SNR is sufficient for all connections Args: state (EnvState): Environment state params (EnvParams): Environment parameters Returns: jnp.array: 1 if SNR is sufficient for connection else 0

Source code in xlron/environments/env_funcs.py 
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
def check_snr_sufficient(state: RSAGNModelEnvState, params: RSAGNModelEnvParams) -> chex.Array:
    """Check if SNR is sufficient for all connections
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: 1 if SNR is sufficient for connection else 0
    """
    # TODO - this check needs to be faster!
    required_snr_array = get_required_snr_se_kurtosis_array(state.modulation_format_index_array, 2, params)
    # Transform lightpath index array by getting lightpath value, getting path-link array, and summing inverse link SNRs
    lightpath_snr_array = get_lightpath_snr(state, params)
    check_snr_sufficient = jnp.where(lightpath_snr_array >= required_snr_array, 0, 1)
    # jax.debug.print("check_snr_sufficient {}", check_snr_sufficient, ordered=True)
    # jax.debug.print("required_snr_array {}", required_snr_array, ordered=True)
    # jax.debug.print("lightpath_snr_array {}", lightpath_snr_array, ordered=True)
    # jax.debug.print("state.modulation_format_index_array {}", state.modulation_format_index_array, ordered=True)
    # jax.debug.print("state.channel_centre_bw_array {}", state.channel_centre_bw_array, ordered=True)
    # jax.debug.print("state.channel_power_array {}", state.channel_power_array, ordered=True)
    return jnp.any(check_snr_sufficient)

check_topology(action_history, topology_pattern)

Check that each unique virtual node (as indicated by topology pattern) is assigned to a consistent physical node i.e. start and end node of ring is same physical node. Method: For each node index in topology pattern, mask action history with that index, then find max value in masked array. If max value is not the same for all values for that virtual node in action history, then return 1, else 0. Array should be all zeroes at the end, so do an any() check on that. e.g. virtual topology pattern = [2,1,3,1,4,1,2] 3 node ring action history = [0,34,4,0,3,1,0] meaning v node "2" goes p node 0, v node "3" goes p node 4, v node "4" goes p node 3 The numbers in-between relate to the slot action. If any value in the array is 1, a virtual node is assigned to multiple different physical nodes. Need to check from both perspectives: 1. For each virtual node, check that all physical nodes are the same 2. For each physical node, check that all virtual nodes are the same

Parameters:

Name Type Description Default
action_history

Action history

 required
topology_pattern

Topology pattern

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
def check_topology(action_history, topology_pattern):
    """Check that each unique virtual node (as indicated by topology pattern) is assigned to a consistent physical node
    i.e. start and end node of ring is same physical node.
    Method:
    For each node index in topology pattern, mask action history with that index, then find max value in masked array.
    If max value is not the same for all values for that virtual node in action history, then return 1, else 0.
    Array should be all zeroes at the end, so do an any() check on that.
    e.g. virtual topology pattern = [2,1,3,1,4,1,2]  3 node ring
    action history = [0,34,4,0,3,1,0]
    meaning v node "2" goes p node 0, v node "3" goes p node 4, v node "4" goes p node 3
    The numbers in-between relate to the slot action.
    If any value in the array is 1, a virtual node is assigned to multiple different physical nodes.
    Need to check from both perspectives:
    1. For each virtual node, check that all physical nodes are the same
    2. For each physical node, check that all virtual nodes are the same

    Args:
        action_history: Action history
        topology_pattern: Topology pattern

    Returns:
        bool: True if check failed, False if check passed
    """
    def loop_func_virtual(i, val):
        # Get indices of physical node in action history that correspond to virtual node i
        masked_val = jnp.where(i == topology_pattern, val, -1)
        # Get maximum value at those indices (should all be same)
        max_node = jnp.max(masked_val)
        # For relevant indices, if max value then return 0 else 1
        val = jnp.where(masked_val != -1, masked_val != max_node, val)
        return val
    def loop_func_physical(i, val):
        # Get indices of virtual nodes in topology pattern that correspond to physical node i
        masked_val = jnp.where(i == action_history, val, -1)
        # Get maximum value at those indices (should all be same)
        max_node = jnp.max(masked_val)
        # For relevant indices, if max value then return 0 else 1
        val = jnp.where(masked_val != -1, masked_val != max_node, val)
        return val
    topology_pattern = topology_pattern[::2]  # Only look at node indices, not slot actions
    action_history = action_history[::2]
    check_virtual = jax.lax.fori_loop(jnp.min(topology_pattern), jnp.max(topology_pattern)+1, loop_func_virtual, action_history)
    check_physical = jax.lax.fori_loop(jnp.min(action_history), jnp.max(action_history)+1, loop_func_physical, topology_pattern)
    check = jnp.concatenate((check_virtual, check_physical))
    return jnp.any(check)

check_unique_nodes(node_departure_array)

Count negative values on each node (row) in node departure array, must not exceed 1.

Parameters:

Name Type Description Default
node_departure_array Array

Node departure array (N x R) where N is number of nodes and R is number of resources

 required

Returns:

Name Type Description
bool bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
@jax.jit
def check_unique_nodes(node_departure_array: chex.Array) -> bool:
    """Count negative values on each node (row) in node departure array, must not exceed 1.

    Args:
        node_departure_array: Node departure array (N x R) where N is number of nodes and R is number of resources

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.any(jnp.sum(jnp.where(node_departure_array < 0, 1, 0), axis=1) > 1)

check_vone_action(state, remaining_actions, total_requested_nodes)

Check if action is valid. Combines checks for: - sufficient node capacities - unique nodes assigned - minimum two nodes assigned - all requested nodes assigned - correct topology pattern - no spectrum reuse

Parameters:

Name Type Description Default
state

current state

 required
remaining_actions

remaining actions

 required
total_requested_nodes

total requested nodes

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
def check_vone_action(state, remaining_actions, total_requested_nodes):
    """Check if action is valid.
    Combines checks for:
    - sufficient node capacities
    - unique nodes assigned
    - minimum two nodes assigned
    - all requested nodes assigned
    - correct topology pattern
    - no spectrum reuse

    Args:
        state: current state
        remaining_actions: remaining actions
        total_requested_nodes: total requested nodes

    Returns:
        bool: True if check failed, False if check passed
    """
    checks = jnp.stack((
        check_node_capacities(state.node_capacity_array),
        check_unique_nodes(state.node_departure_array),
        # TODO (VONE) - Remove two nodes check if impairs performance
        #  (check_all_nodes_assigned is sufficient but fails after last action of request instead of earlier)
        check_min_two_nodes_assigned(state.node_departure_array),
        jax.lax.cond(
            jnp.equal(remaining_actions, jnp.array(1)),
            lambda x: check_all_nodes_assigned(*x),
            lambda x: jnp.array(False),
            (state.node_departure_array, total_requested_nodes)
        ),
        jax.lax.cond(
            jnp.equal(remaining_actions, jnp.array(1)),
            lambda x: check_topology(*x),
            lambda x: jnp.array(False),
            (state.action_history, state.request_array[1])
        ),
        check_no_spectrum_reuse(state.link_slot_array),
    ))
    #jax.debug.print("Checks: {}", checks, ordered=True)
    return jnp.any(checks)

convert_node_probs_to_traffic_matrix(node_probs)

Convert list of node probabilities to symmetric traffic matrix.

Parameters:

Name Type Description Default
node_probs list

node probabilities

 required

Returns:

Name Type Description
traffic_matrix Array

traffic matrix
Source code in xlron/environments/env_funcs.py 
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
def convert_node_probs_to_traffic_matrix(node_probs: list) -> chex.Array:
    """Convert list of node probabilities to symmetric traffic matrix.

    Args:
        node_probs: node probabilities

    Returns:
        traffic_matrix: traffic matrix
    """
    matrix = jnp.outer(node_probs, node_probs)
    # Set lead diagonal to zero
    matrix = jnp.where(jnp.eye(matrix.shape[0]) == 1, 0, matrix)
    matrix = normalise_traffic_matrix(matrix)
    return matrix

create_run_name(config)

Create name for run based on config flags

Source code in xlron/environments/env_funcs.py 
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
def create_run_name(config: Union[box.Box, dict]) -> str:
    """Create name for run based on config flags"""
    env_type = config["env_type"]
    topology = config["topology_name"]
    slots = config["link_resources"]
    gnn = "_GNN" if config["USE_GNN"] else ""
    incremental = "_INC" if config["incremental_loading"] else ""
    run_name = f"{env_type}_{topology}_{slots}{gnn}{incremental}".upper()
    if config["EVAL_HEURISTIC"]:
        run_name += f"_{config['path_heuristic']}"
        if env_type.lower() == "vone":
            run_name += f"_{config['node_heuristic']}"
    elif config["EVAL_MODEL"]:
        run_name += f"_EVAL"
    return run_name

decrement_action_counter(state)

Decrement action counter in-place. Used in VONE environments.

Source code in xlron/environments/env_funcs.py 
547
548
549
550
def decrement_action_counter(state):
    """Decrement action counter in-place. Used in VONE environments."""
    state.action_counter.at[-1].add(-1)
    return state

finalise_action_rsa(state, params)

Turn departure times positive.

Parameters:

Name Type Description Default
state EnvState

current state

 required

Returns:

Name Type Description
state

updated state
Source code in xlron/environments/env_funcs.py 
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
@partial(jax.jit, donate_argnums=(0,))
def finalise_action_rsa(state: EnvState, params: Optional[EnvParams]):
    """Turn departure times positive.

    Args:
        state: current state

    Returns:
        state: updated state
    """
    _, requested_datarate = read_rsa_request(state.request_array)
    state = state.replace(
        link_slot_departure_array=make_positive(state.link_slot_departure_array),
        accepted_services=state.accepted_services + 1,
        accepted_bitrate=state.accepted_bitrate + requested_datarate[0],
        total_bitrate=state.total_bitrate + requested_datarate[0]
    )
    return state

finalise_action_rwalr(state, params)

Turn departure times positive.

Parameters:

Name Type Description Default
state EnvState

current state

 required

Returns:

Name Type Description
state

updated state
Source code in xlron/environments/env_funcs.py 
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
@partial(jax.jit, donate_argnums=(0,))
def finalise_action_rwalr(state: EnvState, params: Optional[EnvParams]):
    """Turn departure times positive.

    Args:
        state: current state

    Returns:
        state: updated state
    """
    _, requested_datarate = read_rsa_request(state.request_array)
    state = state.replace(
        link_slot_departure_array=make_positive(state.link_slot_departure_array),
        accepted_services=state.accepted_services + 1,
        accepted_bitrate=state.accepted_bitrate + requested_datarate[0],
        total_bitrate=state.total_bitrate + requested_datarate[0]
    )
    return state

finalise_vone_action(state)

Turn departure times positive.

Parameters:

Name Type Description Default
state

current state

 required

Returns:

Name Type Description
state

updated state
Source code in xlron/environments/env_funcs.py 
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
@partial(jax.jit, donate_argnums=(0,))
def finalise_vone_action(state):
    """Turn departure times positive.

    Args:
        state: current state

    Returns:
        state: updated state
    """
    state = state.replace(
        node_departure_array=make_positive(state.node_departure_array),
        link_slot_departure_array=make_positive(state.link_slot_departure_array),
        accepted_services=state.accepted_services + 1,
        accepted_bitrate=state.accepted_bitrate  # TODO - get sum of bitrates for requested links
    )
    return state

format_vone_slot_request(state, action)

Format slot request for VONE action into format (source-node, slot, destination-node).

Parameters:

Name Type Description Default
state EnvState

current state

 required
action Array

action to format

 required

Returns:

Type Description
Array

chex.Array: formatted request
Source code in xlron/environments/env_funcs.py 
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
def format_vone_slot_request(state: EnvState, action: chex.Array) -> chex.Array:
    """Format slot request for VONE action into format (source-node, slot, destination-node).

    Args:
        state: current state
        action: action to format

    Returns:
        chex.Array: formatted request
    """
    remaining_actions = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.action_counter, 2, 1))
    full_request = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.request_array, 0, 1))
    unformatted_request = jax.lax.dynamic_slice_in_dim(full_request, (remaining_actions - 1) * 2, 3)
    node_s = jax.lax.dynamic_slice_in_dim(action, 0, 1)
    requested_slots = jax.lax.dynamic_slice_in_dim(unformatted_request, 1, 1)
    node_d = jax.lax.dynamic_slice_in_dim(action, 2, 1)
    formatted_request = jnp.concatenate((node_s, requested_slots, node_d))
    return formatted_request

generate_arrival_holding_times(key, params)

Generate arrival and holding times based on Poisson distributed events. To understand how sampling from e^-x can be transformed to sample from lambdae^-(x/lambda) see: https://en.wikipedia.org/wiki/Inverse_transform_sampling#Examples Basically, inverse transform sampling is used to sample from a distribution with CDF F(x). The CDF of the exponential distribution (lambdae^-{lambdax}) is F(x) = 1 - e^-{lambdax}. Therefore, the inverse CDF is x = -ln(1-u)/lambda, where u is sample from uniform distribution. Therefore, we need to divide jax.random.exponential() by lambda in order to scale the standard exponential CDF. Experimental histograms of this method compared to random.expovariate() in Python's random library show that the two methods are equivalent. Also see: https://numpy.org/doc/stable/reference/random/generated/numpy.random.exponential.html https://jax.readthedocs.io/en/latest/_autosummary/jax.random.exponential.html

Parameters:

Name Type Description Default
key

PRNG key

 required
params

Environment parameters

 required

Returns:

Name Type Description
arrival_time

Arrival time
holding_time

Holding time
Source code in xlron/environments/env_funcs.py 
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
@partial(jax.jit, static_argnums=(1,))
def generate_arrival_holding_times(key, params):
    """
    Generate arrival and holding times based on Poisson distributed events.
    To understand how sampling from e^-x can be transformed to sample from lambda*e^-(x/lambda) see:
    https://en.wikipedia.org/wiki/Inverse_transform_sampling#Examples
    Basically, inverse transform sampling is used to sample from a distribution with CDF F(x).
    The CDF of the exponential distribution (lambda*e^-{lambda*x}) is F(x) = 1 - e^-{lambda*x}.
    Therefore, the inverse CDF is x = -ln(1-u)/lambda, where u is sample from uniform distribution.
    Therefore, we need to divide jax.random.exponential() by lambda in order to scale the standard exponential CDF.
    Experimental histograms of this method compared to random.expovariate() in Python's random library show that
    the two methods are equivalent.
    Also see: https://numpy.org/doc/stable/reference/random/generated/numpy.random.exponential.html
    https://jax.readthedocs.io/en/latest/_autosummary/jax.random.exponential.html

    Args:
        key: PRNG key
        params: Environment parameters

    Returns:
        arrival_time: Arrival time
        holding_time: Holding time
    """
    key_arrival, key_holding = jax.random.split(key, 2)
    arrival_time = jax.random.exponential(key_arrival, shape=(1,)) \
                   / params.arrival_rate  # Divide because it is rate (lambda)
    if params.truncate_holding_time:
        # For DeepRMSA, need to generate holding times that are less than 2*mean_service_holding_time
        key_holding = jax.random.split(key, 5)
        holding_times = jax.vmap(lambda x: jax.random.exponential(x, shape=(1,)) \
                                * params.mean_service_holding_time)(key_holding)
        holding_times = jnp.where(holding_times < 2*params.mean_service_holding_time, holding_times, 0)
        # Get first non-zero value in holding_times
        non_zero_index = jnp.nonzero(holding_times, size=1)[0][0]
        holding_time = jax.lax.dynamic_slice(jnp.squeeze(holding_times), (non_zero_index,), (1,))
    else:
        holding_time = jax.random.exponential(key_holding, shape=(1,)) \
                       * params.mean_service_holding_time  # Multiply because it is mean (1/lambda)
    return arrival_time, holding_time

generate_vone_request(key, state, params)

Generate a new request for the VONE environment. The request has two rows. The first row shows the node and slot values. The first three elements of the second row show the number of unique nodes, the total number of steps, and the remaining steps. These first three elements comprise the action counter. The remaining elements of the second row show the virtual topology pattern, i.e. the connectivity of the virtual topology.

Source code in xlron/environments/env_funcs.py 
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
@partial(jax.jit, static_argnums=(2,))
def generate_vone_request(key: chex.PRNGKey, state: EnvState, params: EnvParams):
    """Generate a new request for the VONE environment.
    The request has two rows. The first row shows the node and slot values.
    The first three elements of the second row show the number of unique nodes, the total number of steps, and the remaining steps.
    These first three elements comprise the action counter.
    The remaining elements of the second row show the virtual topology pattern, i.e. the connectivity of the virtual topology.
    """
    shape = params.max_edges*2+1  # shape of request array
    key_topology, key_node, key_slot, key_times = jax.random.split(key, 4)
    # Randomly select topology, node resources, slot resources
    pattern = jax.random.choice(key_topology, state.virtual_topology_patterns)
    action_counter = jax.lax.dynamic_slice(pattern, (0,), (3,))
    topology_pattern = jax.lax.dynamic_slice(pattern, (3,), (pattern.shape[0]-3,))
    selected_node_values = jax.random.choice(key_node, state.values_nodes, shape=(shape,))
    selected_bw_values = jax.random.choice(key_slot, params.values_bw.val, shape=(shape,))
    # Create a mask for odd and even indices
    mask = jnp.tile(jnp.array([0, 1]), (shape+1) // 2)[:shape]
    # Vectorized conditional replacement using mask
    first_row = jnp.where(mask, selected_bw_values, selected_node_values)
    # Make sure node request values are consistent for same virtual nodes
    first_row = jax.lax.fori_loop(
        2,  # Lowest node index in virtual topology requests is 2
        shape,  # Highest possible node index in virtual topology requests is shape-1
        lambda i, x: jnp.where(topology_pattern == i, selected_node_values[i], x),
        first_row
    )
    # Mask out unused part of request array
    first_row = jnp.where(topology_pattern == 0, 0, first_row)
    # Set times
    arrival_time, holding_time = generate_arrival_holding_times(key, params)
    state = state.replace(
        holding_time=holding_time,
        current_time=state.current_time + arrival_time,
        action_counter=action_counter,
        request_array=jnp.vstack((first_row, topology_pattern)),
        action_history=init_action_history(params),
        total_requests=state.total_requests + 1
    )
    state = remove_expired_node_requests(state) if not params.incremental_loading else state
    state = remove_expired_services_rsa(state) if not params.incremental_loading else state
    return state

get_best_modulation_format(state, path, initial_slot_index, launch_power, params)

Get best modulation format for lightpath. "Best" is the highest order that has SNR requirements below available. Try each modulation format, calculate SNR for each, then return the highest order possible. Args: state (EnvState): Environment state path (chex.Array): Path array initial_slot_index (int): Initial slot index params (EnvParams): Environment parameters Returns: jnp.array: Acceptable modulation format indices

Source code in xlron/environments/env_funcs.py 
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
@partial(jax.jit, static_argnums=(3,))
def get_best_modulation_format(state: EnvState, path: chex.Array, initial_slot_index: int, launch_power: chex.Array, params: EnvParams) -> chex.Array:
    """Get best modulation format for lightpath. "Best" is the highest order that has SNR requirements below available.
    Try each modulation format, calculate SNR for each, then return the highest order possible.
    Args:
        state (EnvState): Environment state
        path (chex.Array): Path array
        initial_slot_index (int): Initial slot index
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Acceptable modulation format indices
    """
    _, requested_datarate = read_rsa_request(state.request_array)
    mod_format_count = params.modulations_array.val.shape[0]
    acceptable_mod_format_indices = jnp.full((mod_format_count,), -2)

    def acceptable_modulation_format(i, acceptable_format_indices):
        req_snr = params.modulations_array.val[i][2] + params.snr_margin
        se = params.modulations_array.val[i][1]
        req_slots = required_slots(requested_datarate, se, params.slot_size, params.guardband)
        # TODO - need to check we don't overwrite values in already occupied slots
        # Possible approaches:
        # Check slot occupancy? Probably would need to iterate through for num_slots, but that's an issue
        # What about we allocate and then fix up later, e.g. could it be possible to just add the modulation format on top without
        # check sum of path links prior to assigning?
        #
        new_state = state.replace(
            channel_power_array=vmap_set_path_links(
                state.channel_power_array, path, initial_slot_index, req_slots, launch_power),
            channel_centre_bw_array=vmap_set_path_links(
                state.channel_centre_bw_array, path, initial_slot_index, req_slots, params.slot_size)
        )
        snr_value = get_minimum_snr_of_channels_on_path(new_state, path, initial_slot_index, req_slots, params)
        # jax.debug.print("snr_value {}", snr_value, ordered=True)
        # jax.debug.print("req_snr {}", req_snr, ordered=True)
        acceptable_format_index = jnp.where(snr_value >= req_snr, i, -1).reshape((1,))
        acceptable_format_indices = jax.lax.dynamic_update_slice(acceptable_format_indices, acceptable_format_index, (i,))
        # jax.debug.print("acceptable_format_indices {}", acceptable_format_indices, ordered=True)
        return acceptable_format_indices

    acceptable_mod_format_indices = jax.lax.fori_loop(
        0,
        mod_format_count,
        acceptable_modulation_format,
        acceptable_mod_format_indices
    )
    return acceptable_mod_format_indices

get_best_modulation_format_simple(state, path, initial_slot_index, params)

Get modulation format for lightpath. Assume worst case (least Gaussian) modulation format when calculating SNR. Args: state (EnvState): Environment state path (chex.Array): Path array initial_slot_index (int): Initial slot index params (EnvParams): Environment parameters Returns: jnp.array: Acceptable modulation format indices

Source code in xlron/environments/env_funcs.py 
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
@partial(jax.jit, static_argnums=(3,))
def get_best_modulation_format_simple(
        state: RSAGNModelEnvState, path: chex.Array, initial_slot_index: int, params: RSAGNModelEnvParams
) -> chex.Array:
    """Get modulation format for lightpath.
    Assume worst case (least Gaussian) modulation format when calculating SNR.
    Args:
        state (EnvState): Environment state
        path (chex.Array): Path array
        initial_slot_index (int): Initial slot index
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Acceptable modulation format indices
    """
    link_snr_array = get_snr_link_array(state, params)
    snr_value = get_snr_for_path(path, link_snr_array, params)[initial_slot_index] - params.snr_margin  # Margin
    mod_format_count = params.modulations_array.val.shape[0]
    acceptable_mod_format_indices = jnp.arange(mod_format_count)
    req_snr = params.modulations_array.val[:, 2] + params.snr_margin
    acceptable_mod_format_indices = jnp.where(snr_value >= req_snr,
                                              acceptable_mod_format_indices,
                                              jnp.full((mod_format_count,), -2))
    return acceptable_mod_format_indices

get_centre_frequency(initial_slot_index, num_slots, params)

Get centre frequency for new lightpath

Parameters:

Name Type Description Default
initial_slot_index Array

Centre frequency of first slot

 required
num_slots float

Number of slots

 required
params RSAGNModelEnvParams

Environment parameters

 required

Returns:

Type Description
Array

chex.Array: Centre frequency for new lightpath
Source code in xlron/environments/env_funcs.py 
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
@partial(jax.jit, static_argnums=(2,))
def get_centre_frequency(initial_slot_index: int, num_slots: int, params: RSAGNModelEnvParams) -> chex.Array:
    """Get centre frequency for new lightpath

    Args:
        initial_slot_index (chex.Array): Centre frequency of first slot
        num_slots (float): Number of slots
        params (RSAGNModelEnvParams): Environment parameters

    Returns:
        chex.Array: Centre frequency for new lightpath
    """
    slot_centres = (jnp.arange(params.link_resources) - (params.link_resources + 1) / 2) * params.slot_size
    return slot_centres[initial_slot_index] + (params.slot_size * (num_slots + 1)) / 2

get_edge_disjoint_paths(graph)

Get edge disjoint paths between all nodes in graph.

Parameters:

Name Type Description Default
graph Graph

graph

 required

Returns:

Name Type Description
dict dict

edge disjoint paths (path is list of edges)
Source code in xlron/environments/env_funcs.py 
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
def get_edge_disjoint_paths(graph: nx.Graph) -> dict:
    """Get edge disjoint paths between all nodes in graph.

    Args:
        graph: graph

    Returns:
        dict: edge disjoint paths (path is list of edges)
    """
    result = {n: {} for n in graph}
    for n1, n2 in itertools.combinations(graph, 2):
        # Sort by number of links in path
        # TODO - sort by path length
        result[n1][n2] = sorted(list(nx.edge_disjoint_paths(graph, n1, n2)), key=len)
        result[n2][n1] = sorted(list(nx.edge_disjoint_paths(graph, n2, n1)), key=len)
    return result

get_launch_power(state, path_action, power_action, params)

Get launch power for new lightpath. N.B. launch power is specified in dBm but is converted to linear units when stored in channel_power_array. This func returns linear units (mW). Path action is used to determine the launch power in the case of tabular launch power type. Power action is used to determine the launch power in the case of RL launch power type. During masking, power action is set as state.launch_power_array[0], which is set by the RL agent. Args: state (EnvState): Environment state path_action (chex.Array): Action specifying path index (0 to k_paths-1) power_action (chex.Array): Action specifying launch power in dBm params (EnvParams): Environment parameters Returns: chex.Array: Launch power for new lightpath

Source code in xlron/environments/env_funcs.py 
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
@partial(jax.jit, static_argnums=(1,))
def get_launch_power(state: EnvState, path_action: chex.Array, power_action: chex.Array, params: EnvParams) -> chex.Array:
    """Get launch power for new lightpath. N.B. launch power is specified in dBm but is converted to linear units
    when stored in channel_power_array. This func returns linear units (mW).
    Path action is used to determine the launch power in the case of tabular launch power type.
    Power action is used to determine the launch power in the case of RL launch power type. During masking,
    power action is set as state.launch_power_array[0], which is set by the RL agent.
    Args:
        state (EnvState): Environment state
        path_action (chex.Array): Action specifying path index (0 to k_paths-1)
        power_action (chex.Array): Action specifying launch power in dBm
        params (EnvParams): Environment parameters
    Returns:
        chex.Array: Launch power for new lightpath
    """
    k_path_index, _ = process_path_action(state, params, path_action)
    if params.launch_power_type == 1:  # Fixed
        return state.launch_power_array[0]
    elif params.launch_power_type == 2:  # Tabular (one row per path)
        nodes_sd, requested_datarate = read_rsa_request(state.request_array)
        source, dest = nodes_sd
        i = get_path_indices(source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph).astype(
            jnp.int32)
        return state.launch_power_array[i+k_path_index]
    elif params.launch_power_type == 3:  # RL
        return power_action
    elif params.launch_power_type == 4:  # Scaled
        nodes_sd, requested_datarate = read_rsa_request(state.request_array)
        source, dest = nodes_sd
        i = get_path_indices(
            source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph
        ).astype(jnp.int32)
        # Get path length
        link_length_array = jnp.sum(params.link_length_array.val, axis=1)
        path_length = jnp.sum(link_length_array[i+k_path_index])
        maximum_path_length = jnp.max(jnp.dot(params.path_link_array.val, params.link_length_array.val))
        return state.launch_power_array[0] * (path_length / maximum_path_length)
    else:
        raise ValueError("Invalid launch power type. Check params.launch_power_type")

get_lightpath_snr(state, params)

Get SNR for each link on path. N.B. that in most cases it is more efficient to calculate the SNR for every possible path, rather than a slot-by-slot basis. But in some cases slot-by-slot is better i.e. when kN(N-1)/2 > LS Args: state (RSAGNModelEnvState): Environment state params (RSAGNModelEnvParams): Environment parameters

Returns:

Type Description
Array

chex.array: SNR for each link on path
Source code in xlron/environments/env_funcs.py 
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
def get_lightpath_snr(state: RSAGNModelEnvParams, params: RSAGNModelEnvParams) -> chex.Array:
    """Get SNR for each link on path.
    N.B. that in most cases it is more efficient to calculate the SNR for every possible path, rather than a slot-by-slot basis.
    But in some cases slot-by-slot is better i.e. when k*N(N-1)/2 > L*S
    Args:
        state (RSAGNModelEnvState): Environment state
        params (RSAGNModelEnvParams): Environment parameters

    Returns:
        chex.array: SNR for each link on path
    """
    # Get the SNR for the channel that the path occupies
    path_snr_array = jax.vmap(get_snr_for_path, in_axes=(0, None, None))(params.path_link_array.val, state.link_snr_array, params)
    # Where value in path_index_array matches index of path_snr_array, substitute in SNR value
    slot_indices = jnp.arange(params.link_resources)
    lightpath_snr_array = jax.vmap(jax.vmap(lambda x, si: path_snr_array[x][si], in_axes=(0, 0)), in_axes=(0, None))(state.path_index_array, slot_indices)
    return lightpath_snr_array

get_minimum_snr_of_channels_on_path(state, path, slot_index, req_slots, params)

Get the minimum value of the SNR on newly assigned channels. N.B. this requires the link_snr_array to have already been calculated and present in state.

Source code in xlron/environments/env_funcs.py 
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
@partial(jax.jit, static_argnums=(2,))
def get_minimum_snr_of_channels_on_path(
        state: RSAGNModelEnvState, path: chex.Array, slot_index: chex.Array, req_slots: int, params: RSAGNModelEnvParams
) -> chex.Array:
    """Get the minimum value of the SNR on newly assigned channels.
    N.B. this requires the link_snr_array to have already been calculated and present in state."""
    snr_value_all_channels = get_snr_for_path(path, state.link_snr_array, params)
    min_snr_value_sub_channels = jnp.min(
        jnp.concatenate([
            snr_value_all_channels[slot_index].reshape((1,)),
            snr_value_all_channels[slot_index + req_slots - 1].reshape((1,))
        ], axis=0)
    )
    return min_snr_value_sub_channels

get_num_spectral_features(n_nodes)

Heuristic for number of spectral features based on graph size.

Parameters:

Name Type Description Default
n_nodes int

Number of nodes in the graph

 required

Returns:

Type Description
int

Number of spectral features to use, clamped between 3 and 15.
int

Follows log2(n_nodes) scaling as reasonable default.
Source code in xlron/environments/env_funcs.py 
35
36
37
38
39
40
41
42
43
44
45
46
@partial(jax.jit, static_argnums=(0,))
def get_num_spectral_features(n_nodes: int) -> int:
    """Heuristic for number of spectral features based on graph size.

    Args:
        n_nodes: Number of nodes in the graph

    Returns:
        Number of spectral features to use, clamped between 3 and 15.
        Follows log2(n_nodes) scaling as reasonable default.
    """
    return jnp.minimum(jnp.maximum(3, jnp.floor(jnp.log2(n_nodes))), 15).astype(int)

get_path_from_path_index_array(path_index_array, path_link_array)

Get path from path index array. Args: path_index_array (chex.Array): Path index array path_link_array (chex.Array): Path link array

Returns:

Type Description
Array

jnp.array: path index values replaced with binary path-link arrays
Source code in xlron/environments/env_funcs.py 
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
@partial(jax.jit, static_argnums=(1,))
def get_path_from_path_index_array(path_index_array: chex.Array, path_link_array: chex.Array) -> chex.Array:
    """Get path from path index array.
    Args:
        path_index_array (chex.Array): Path index array
        path_link_array (chex.Array): Path link array

    Returns:
        jnp.array: path index values replaced with binary path-link arrays
    """
    def get_index_from_link(link):
        return jax.vmap(lambda x: path_link_array[x], in_axes=(0,))(link)

    return jax.vmap(get_index_from_link, in_axes=(0,))(path_index_array)

get_path_index_array(params, nodes)

Indices of paths between source and destination from path array

Source code in xlron/environments/env_funcs.py 
742
743
744
745
746
747
748
749
@partial(jax.jit, static_argnums=(0,))
def get_path_index_array(params, nodes):
    """Indices of paths between source and destination from path array"""
    # get source and destination nodes in order (for accurate indexing of path-link array)
    source, dest = nodes
    i = get_path_indices(source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph).astype(jnp.int32)
    index_array = jax.lax.dynamic_slice(jnp.arange(0, params.path_link_array.shape[0]), (i,), (params.k_paths,))
    return index_array

get_path_indices(s, d, k, N, directed=False)

Get path indices for a given source, destination and number of paths. If source > destination and the graph is directed (two fibres per link, one in each direction) then an offset is added to the index to get the path in the other direction (the offset is the total number source-dest pairs).

Parameters:

Name Type Description Default
s int

Source node index

 required
d int

Destination node index

 required
k int

Number of paths

 required
N int

Number of nodes

 required
directed bool

Whether graph is directed. Defaults to False.

 False

Returns:

Type Description
Array

jnp.array: Start index on path-link array for candidate paths
Source code in xlron/environments/env_funcs.py 
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
@partial(jax.jit, static_argnums=(2, 3, 4))
def get_path_indices(s: int, d: int, k: int, N: int, directed: bool = False) -> chex.Array:
    """Get path indices for a given source, destination and number of paths.
    If source > destination and the graph is directed (two fibres per link, one in each direction) then an offset is
    added to the index to get the path in the other direction (the offset is the total number source-dest pairs).

    Args:
        s (int): Source node index
        d (int): Destination node index
        k (int): Number of paths
        N (int): Number of nodes
        directed (bool, optional): Whether graph is directed. Defaults to False.

    Returns:
        jnp.array: Start index on path-link array for candidate paths
    """
    node_indices = jnp.arange(N, dtype=jnp.int32)
    indices_to_s = jnp.where(node_indices < s, node_indices, 0)
    indices_to_d = jnp.where(node_indices < d, node_indices, 0)
    # If two fibres per link, add offset to index to get fibre in other direction if source > destination
    directed_offset = directed * (s > d) * N * (N - 1) * k / 2
    # The following equation is based on the combinations formula
    forward = ((N * s + d - jnp.sum(indices_to_s) - 2 * s - 1) * k)
    backward = ((N * d + s - jnp.sum(indices_to_d) - 2 * d - 1) * k)
    return forward * (s < d) + backward * (s > d) + directed_offset

get_path_slots(link_slot_array, params, nodes_sd, i, agg_func='max')

Get slots on each constitutent link of path from link_slot_array (L x S), then aggregate to get (S x 1) representation of slots on path.

Parameters:

Name Type Description Default
link_slot_array Array

link-slot array

 required
params EnvParams

environment parameters

 required
nodes_sd Array

source-destination nodes

 required
i int

path index

 required
agg_func str

aggregation function (max or sum). If max, result will be available slots on path. If sum, result will contain information on edge features.

 'max'

Returns:

Name Type Description
slots Array

slots on path
Source code in xlron/environments/env_funcs.py 
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
@partial(jax.jit, static_argnums=(1, 4))
def get_path_slots(link_slot_array: chex.Array, params: EnvParams, nodes_sd: chex.Array, i: int, agg_func: str = "max") -> chex.Array:
    """Get slots on each constitutent link of path from link_slot_array (L x S),
    then aggregate to get (S x 1) representation of slots on path.

    Args:
        link_slot_array: link-slot array
        params: environment parameters
        nodes_sd: source-destination nodes
        i: path index
        agg_func: aggregation function (max or sum).
            If max, result will be available slots on path.
            If sum, result will contain information on edge features.

    Returns:
        slots: slots on path
    """
    path = get_paths(params, nodes_sd)[i]
    path = path.reshape((params.num_links, 1))
    # Get links and collapse to single dimension
    num_slots = params.link_resources if agg_func == "max" else math.ceil(params.link_resources/params.aggregate_slots)
    slots = jnp.where(path, link_slot_array, jnp.zeros(num_slots))
    # Make any -1s positive then get max for each slot across links
    if agg_func == "max":
        # Use this for getting slots from link_slot_array
        slots = jnp.max(jnp.absolute(slots), axis=0)
    elif agg_func == "sum":
        # TODO - consider using an RNN (or S5) to aggregate edge features
        # Use this (or alternative) for aggregating edge features from GNN
        slots = jnp.sum(slots, axis=0)
    else:
        raise ValueError("agg_func must be 'max' or 'sum'")
    return slots

get_paths(params, nodes)

Get k paths between source and destination

Source code in xlron/environments/env_funcs.py 
752
753
754
755
756
@partial(jax.jit, static_argnums=(0,))
def get_paths(params, nodes):
    """Get k paths between source and destination"""
    index_array = get_path_index_array(params, nodes)
    return jnp.take(params.path_link_array.val, index_array, axis=0)

get_paths_obs_gn_model(state, params)

Get observation space for launch power optimization (with numerical stability).

Source code in xlron/environments/env_funcs.py 
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
@partial(jax.jit, static_argnums=(1,))
def get_paths_obs_gn_model(state: RSAGNModelEnvState, params: RSAGNModelEnvParams) -> chex.Array:
    # TODO - make this just show the stats from just one path at a time
    """Get observation space for launch power optimization (with numerical stability)."""
    request_array = state.request_array.reshape((-1,))
    path_stats = calculate_path_stats(state, params, request_array)
    # Remove first 3 items of path stats for each path
    path_stats = path_stats[:, 3:]
    link_length_array = jnp.sum(params.link_length_array.val, axis=1)
    lightpath_snr_array = get_lightpath_snr(state, params)
    nodes_sd, requested_datarate = read_rsa_request(request_array)
    source, dest = nodes_sd

    def calculate_gn_path_stats(k_path_index, init_val):
        # Get path index
        path_index = get_path_indices(source, dest, params.k_paths, params.num_nodes,
                                      directed=params.directed_graph).astype(
            jnp.int32) + k_path_index
        path = params.path_link_array[path_index]
        path_length = jnp.dot(path, link_length_array)
        max_path_length = jnp.max(jnp.dot(params.path_link_array.val, link_length_array))
        path_length_norm = path_length / max_path_length
        path_length_hops_norm = jnp.sum(path) / jnp.max(jnp.sum(params.path_link_array.val, axis=1))
        # Connections on path
        num_connections = jnp.where(path == 1, jnp.where(state.channel_power_array > 0, 1, 0).sum(axis=1), 0).sum()
        num_connections_norm = num_connections / params.link_resources
        # Mean power of connections on path
        # make path with row length equal to link_resource (+1 to avoid zero division)
        mean_power_norm = (jnp.where(path == 1, state.channel_power_array.sum(axis=1), 0).sum() /
                           (jnp.where(num_connections > 0., num_connections, 1.) * params.max_power))
        # Mean SNR of connections on the path links
        max_snr = 50.  # Nominal value for max GSNR in dB
        mean_snr_norm = (jnp.where(path == 1, lightpath_snr_array.sum(axis=1), 0).sum() /
                         (jnp.where(num_connections > 0., num_connections, 1.) * max_snr))
        return jax.lax.dynamic_update_slice(
            init_val,
            jnp.array([[
                path_length,
                path_length_hops_norm,
                num_connections_norm,
                mean_power_norm,
                mean_snr_norm
            ]]),
            (k_path_index, 0),
        )

    gn_path_stats = jnp.zeros((params.k_paths, 5))
    gn_path_stats = jax.lax.fori_loop(
        0, params.k_paths, calculate_gn_path_stats, gn_path_stats
    )
    all_stats = jnp.concatenate([path_stats, gn_path_stats], axis=1)
    return jnp.concatenate(
        (
            jnp.array([source]),
            requested_datarate / 100.,
            jnp.array([dest]),
            jnp.reshape(state.holding_time, (-1,)),
            jnp.reshape(all_stats, (-1,)),
        ),
        axis=0,
    )

get_paths_se(params, nodes)

Get max. spectral efficiency of modulation format on k paths between source and destination

Source code in xlron/environments/env_funcs.py 
759
760
761
762
763
764
@partial(jax.jit, static_argnums=(0,))
def get_paths_se(params, nodes):
    """Get max. spectral efficiency of modulation format on k paths between source and destination"""
    # get source and destination nodes in order (for accurate indexing of path-link array)
    index_array = get_path_index_array(params, nodes)
    return jnp.take(params.path_se_array.val, index_array, axis=0)

get_required_snr_se_kurtosis_array(modulation_format_index_array, col_index, params)

Convert modulation format index to required SNR or spectral efficiency. Modulation format index array contains the index of the modulation format used by the channel. The modulation index references a row in the modulations array, which contains SNR and SE values.

Parameters:

Name Type Description Default
modulation_format_index_array Array

Modulation format index array

 required
col_index int

Column index for required SNR or spectral efficiency

 required
params RSAGNModelEnvParams

Environment parameters

 required

Returns:

Type Description
Array

jnp.array: Required SNR for each channel (min. SNR for empty channel (mod. index 0))
Source code in xlron/environments/env_funcs.py 
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
@partial(jax.jit, static_argnums=(1, 2,))
def get_required_snr_se_kurtosis_array(modulation_format_index_array: chex.Array, col_index: int, params: RSAGNModelEnvParams) -> chex.Array:
    """Convert modulation format index to required SNR or spectral efficiency.
    Modulation format index array contains the index of the modulation format used by the channel.
    The modulation index references a row in the modulations array, which contains SNR and SE values.

    Args:
        modulation_format_index_array (chex.Array): Modulation format index array
        col_index (int): Column index for required SNR or spectral efficiency
        params (RSAGNModelEnvParams): Environment parameters

    Returns:
        jnp.array: Required SNR for each channel (min. SNR for empty channel (mod. index 0))
    """
    return jax.vmap(get_required_snr_se_kurtosis_on_link, in_axes=(0, None, None))(modulation_format_index_array, col_index, params)

Get SNR per link Args: state (EnvState): Environment state params (EnvParams): Environment parameters Returns: jnp.array: SNR per link

Source code in xlron/environments/env_funcs.py 
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
@partial(jax.jit, static_argnums=(1,))
def get_snr_link_array(state: EnvState, params: EnvParams) -> chex.Array:
    """Get SNR per link
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: SNR per link
    """

    def get_link_snr(link_index, state, params):
        # Get channel power, channel centre, bandwidth, and noise figure
        link_lengths = params.link_length_array[link_index, :]
        num_spans = jnp.ceil(jnp.sum(link_lengths)*1e3 / params.max_span_length).astype(jnp.int32)
        if params.mod_format_correction:
            mod_format_link = state.modulation_format_index_array[link_index, :]
            kurtosis_link = get_required_snr_se_kurtosis_on_link(mod_format_link, 4, params)
            se_link = get_required_snr_se_kurtosis_on_link(mod_format_link, 1, params)
        else:
            kurtosis_link = jnp.zeros(params.link_resources)
            se_link = jnp.ones(params.link_resources)
        bw_link = state.channel_centre_bw_array[link_index, :]
        ch_power_link = state.channel_power_array[link_index, :]
        required_slots_link = get_required_slots_on_link(bw_link, se_link, params)
        ch_centres_link = get_centre_freq_on_link(jnp.arange(params.link_resources), required_slots_link, params)

        # Calculate SNR
        P = dict(
            num_channels=params.link_resources,
            num_spans=num_spans,
            max_spans=params.max_spans,
            ref_lambda=params.ref_lambda,
            length=link_lengths,
            attenuation_i=jnp.array([params.attenuation]),
            attenuation_bar_i=jnp.array([params.attenuation_bar]),
            nonlinear_coeff=jnp.array([params.nonlinear_coeff]),
            raman_gain_slope_i=jnp.array([params.raman_gain_slope]),
            dispersion_coeff=jnp.array([params.dispersion_coeff]),
            dispersion_slope=jnp.array([params.dispersion_slope]),
            coherent=params.coherent,
            num_roadms=params.num_roadms,
            roadm_loss=params.roadm_loss,
            noise_figure=params.noise_figure,  # TODO (GN MODEL) - support separate noise figures for C and L bands (4 and 6 dB)
            mod_format_correction=params.mod_format_correction,
            ch_power_w_i=ch_power_link.reshape((params.link_resources, 1)),
            ch_centre_i=ch_centres_link.reshape((params.link_resources, 1))*1e9,
            ch_bandwidth_i=bw_link.reshape((params.link_resources, 1))*1e9,
            excess_kurtosis_i=kurtosis_link.reshape((params.link_resources, 1)),
        )
        snr = isrs_gn_model.get_snr(**P)[0]

        return snr

    link_snr_array = jax.vmap(get_link_snr, in_axes=(0, None, None))(jnp.arange(params.num_links), state, params)
    link_snr_array = jnp.nan_to_num(link_snr_array, nan=1e-5)
    return link_snr_array

get_spectral_features(laplacian, num_features)

Compute spectral node features from symmetric normalized graph Laplacian.

Parameters:

Name Type Description Default
adj

Adjacency matrix of the graph

 required
num_features int

Number of eigenvector features to extract

 required

Returns:

Type Description
ndarray

Array of shape (n_nodes, num_features) containing eigenvectors corresponding
ndarray

to the smallest non-zero eigenvalues of the graph Laplacian.
Notes
  • Skips trivial eigenvectors (those with near-zero eigenvalues)
  • Eigenvectors are ordered by ascending eigenvalue magnitude
  • Runtime is O(n^3) - use only for small/medium graphs
  • Eigenvector signs are arbitrary (may vary between runs)
Source code in xlron/environments/env_funcs.py 
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
def get_spectral_features(laplacian: jnp.array, num_features: int) -> jnp.ndarray:
    """Compute spectral node features from symmetric normalized graph Laplacian.

    Args:
        adj: Adjacency matrix of the graph
        num_features: Number of eigenvector features to extract

    Returns:
        Array of shape (n_nodes, num_features) containing eigenvectors corresponding
        to the smallest non-zero eigenvalues of the graph Laplacian.

    Notes:
        - Skips trivial eigenvectors (those with near-zero eigenvalues)
        - Eigenvectors are ordered by ascending eigenvalue magnitude
        - Runtime is O(n^3) - use only for small/medium graphs
        - Eigenvector signs are arbitrary (may vary between runs)
    """
    n_nodes = laplacian.shape[0]
    eigenvalues, eigenvectors = jnp.linalg.eigh(laplacian)
    # Skip trivial eigenvectors (zero eigenvalues for connected components)
    num_zero = jnp.sum(jnp.abs(eigenvalues) < 1e-5)
    valid_features = jnp.minimum(num_features, n_nodes - num_zero)
    return eigenvectors[:, :num_features]#num_zero:num_zero + valid_features]

implement_action_rmsa_gn_model(state, action, params)

Implement action for RSA GN model. Update following arrays: - link_slot_array - link_slot_departure_array - link_snr_array - modulation_format_index_array - channel_power_array - active_path_array Args: state (EnvState): Environment state action (chex.Array): Action tuple (first is path action, second is launch_power) params (EnvParams): Environment parameters Returns: EnvState: Updated environment state

Source code in xlron/environments/env_funcs.py 
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
@partial(jax.jit, static_argnums=(2,))
def implement_action_rmsa_gn_model(
        state: RSAGNModelEnvState, action: chex.Array, params: RSAGNModelEnvParams
) -> EnvState:
    """Implement action for RSA GN model. Update following arrays:
    - link_slot_array
    - link_slot_departure_array
    - link_snr_array
    - modulation_format_index_array
    - channel_power_array
    - active_path_array
    Args:
        state (EnvState): Environment state
        action (chex.Array): Action tuple (first is path action, second is launch_power)
        params (EnvParams): Environment parameters
    Returns:
        EnvState: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_action, power_action = action
    path_action = path_action.astype(jnp.int32)
    k_path_index, initial_slot_index = process_path_action(state, params, path_action)
    lightpath_index = get_lightpath_index(params, nodes_sd, k_path_index)
    path = get_paths(params, nodes_sd)[k_path_index]
    launch_power = get_launch_power(state, path_action, power_action, params)
    # TODO(GN MODEL) - get mod. format based on maximum reach
    mod_format_index = jax.lax.dynamic_slice(
        state.mod_format_mask, (path_action,), (1,)
    ).astype(jnp.int32)[0]
    se = params.modulations_array.val[mod_format_index][1]
    num_slots = required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband)
    # Update link_slot_array and link_slot_departure_array, then other arrays
    state = implement_path_action(state, path, initial_slot_index, num_slots)
    state = state.replace(
        path_index_array=vmap_set_path_links(state.path_index_array, path, initial_slot_index, num_slots, lightpath_index),
        channel_power_array=vmap_set_path_links(state.channel_power_array, path, initial_slot_index, num_slots, launch_power),
        modulation_format_index_array=vmap_set_path_links(state.modulation_format_index_array, path, initial_slot_index, num_slots, mod_format_index),
        channel_centre_bw_array=vmap_set_path_links(state.channel_centre_bw_array, path, initial_slot_index, num_slots, params.slot_size),
    )
    # Update link_snr_array
    state = state.replace(link_snr_array=get_snr_link_array(state, params))
    # jax.debug.print("launch_power {}", launch_power, ordered=True)
    # jax.debug.print("mod_format_index {}", mod_format_index, ordered=True)
    # jax.debug.print("initial_slot_index {}", initial_slot_index, ordered=True)
    # jax.debug.print("state.mod_format_mask {}", state.mod_format_mask, ordered=True)
    # jax.debug.print("path_snr {}", get_snr_for_path(path, state.link_snr_array, params), ordered=True)
    # jax.debug.print("required_snr {}", params.modulations_array.val[mod_format_index][2] + params.snr_margin, ordered=True)
    return state

implement_action_rsa(state, action, params)

Implement action to assign slots on links.

Parameters:

Name Type Description Default
state EnvState

current state

 required
action Array

action to implement

 required
params EnvParams

environment parameters

 required

Returns:

Name Type Description
state EnvState

updated state
Source code in xlron/environments/env_funcs.py 
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
@partial(jax.jit, static_argnums=(2,))
def implement_action_rsa(
        state: EnvState,
        action: chex.Array,
        params: EnvParams,
) -> EnvState:
    """Implement action to assign slots on links.

    Args:
        state: current state
        action: action to implement
        params: environment parameters

    Returns:
        state: updated state
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_index, initial_slot_index = process_path_action(state, params, action)
    path = get_paths(params, nodes_sd)[path_index]
    if params.__class__.__name__ == "RWALightpathReuseEnvParams":
        state = state.replace(
            link_capacity_array=vmap_update_path_links(
                state.link_capacity_array, path, initial_slot_index, 1, requested_datarate
            )
        )
        # TODO (Dynamic-RWALR) - to support diverse requested_datarates for RWA-LR, need to update masking
        # TODO (Dynamic-RWALR) - In order to enable dynamic RWA with lightpath reuse (as opposed to just incremental loading),
        #  need to keep track of active requests OR just randomly remove connections
        #  (could do this by using the link_slot_departure array in a novel way... i.e. don't fill it with departure time but current bw)
        capacity_mask = jnp.where(state.link_capacity_array <= 0., -1., 0.)
        over_capacity_mask = jnp.where(state.link_capacity_array < 0., -1., 0.)
        total_mask = capacity_mask + over_capacity_mask
        state = state.replace(
            link_slot_array=total_mask,
            link_slot_departure_array=vmap_update_path_links(state.link_slot_departure_array, path,
                                                                       initial_slot_index, 1,
                                                                       state.current_time + state.holding_time)
        )
    else:
        se = get_paths_se(params, nodes_sd)[path_index] if params.consider_modulation_format else 1
        num_slots = required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband)
        state = implement_path_action(state, path, initial_slot_index, num_slots)
    return state

implement_action_rsa_gn_model(state, action, params)

Implement action for RSA GN model. Update following arrays: - link_slot_array - link_slot_departure_array - link_snr_array - modulation_format_index_array - channel_power_array - active_path_array Args: state (EnvState): Environment state action (chex.Array): Action tuple (first is path action, second is launch_power) params (EnvParams): Environment parameters Returns: EnvState: Updated environment state

Source code in xlron/environments/env_funcs.py 
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
@partial(jax.jit, static_argnums=(2,))
def implement_action_rsa_gn_model(
        state: RSAGNModelEnvState, action: chex.Array, params: RSAGNModelEnvParams
) -> EnvState:
    """Implement action for RSA GN model. Update following arrays:
    - link_slot_array
    - link_slot_departure_array
    - link_snr_array
    - modulation_format_index_array
    - channel_power_array
    - active_path_array
    Args:
        state (EnvState): Environment state
        action (chex.Array): Action tuple (first is path action, second is launch_power)
        params (EnvParams): Environment parameters
    Returns:
        EnvState: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_action, power_action = action
    path_action = path_action.astype(jnp.int32)
    k_path_index, initial_slot_index = process_path_action(state, params, path_action)
    lightpath_index = get_lightpath_index(params, nodes_sd, k_path_index)
    path = get_paths(params, nodes_sd)[k_path_index]
    launch_power = get_launch_power(state, path_action, power_action, params)
    num_slots = required_slots(requested_datarate, 1, params.slot_size, guardband=params.guardband)
    # Update link_slot_array and link_slot_departure_array, then other arrays
    state = implement_path_action(state, path, initial_slot_index, num_slots)
    state = state.replace(
        path_index_array=vmap_set_path_links(state.path_index_array, path, initial_slot_index, num_slots, lightpath_index),
        channel_power_array=vmap_set_path_links(state.channel_power_array, path, initial_slot_index, num_slots, launch_power),
        channel_centre_bw_array=vmap_set_path_links(state.channel_centre_bw_array, path, initial_slot_index, num_slots, params.slot_size),
        active_lightpaths_array=update_active_lightpaths_array(state, lightpath_index),
        active_lightpaths_array_departure=update_active_lightpaths_array_departure(state, -state.current_time-state.holding_time),
    )
    # Update link_snr_array
    state = state.replace(link_snr_array=get_snr_link_array(state, params))
    return state

implement_action_rwalr(state, action, params)

For use in RWALightpathReuseEnv. Update link_slot_array and link_slot_departure_array to reflect new lightpath assignment. Update link_capacity_array with new capacity if lightpath is available. Undo link_capacity_update if over capacity.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
action Array

Action array

 required
params EnvParams

Environment parameters

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
@partial(jax.jit, static_argnums=(2,))
def implement_action_rwalr(state: EnvState, action: chex.Array, params: EnvParams) -> EnvState:
    """For use in RWALightpathReuseEnv.
    Update link_slot_array and link_slot_departure_array to reflect new lightpath assignment.
    Update link_capacity_array with new capacity if lightpath is available.
    Undo link_capacity_update if over capacity.

    Args:
        state: Environment state
        action: Action array
        params: Environment parameters

    Returns:
        state: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_index, initial_slot_index = process_path_action(state, params, action)
    path = get_paths(params, nodes_sd)[path_index]
    lightpath_available_check, lightpath_existing_check, curr_lightpath_capacity, lightpath_index = (
        check_lightpath_available_and_existing(state, params, action)
    )
    # Get path capacity - request
    lightpath_capacity = jax.lax.cond(
        lightpath_existing_check,
        lambda x: curr_lightpath_capacity - requested_datarate,  # Subtract requested_datarate from current lightpath
        lambda x: jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.path_capacity_array, x, 1)) - requested_datarate,  # Get initial capacity of lightpath - request
        lightpath_index
    )
    # Update link_capacity_array with new capacity if lightpath is available
    state = jax.lax.cond(
        lightpath_available_check,
        lambda x: x.replace(
            link_capacity_array=vmap_set_path_links(
                state.link_capacity_array, path, initial_slot_index, 1, lightpath_capacity
            ),
            path_index_array=vmap_set_path_links(
                state.path_index_array, path, initial_slot_index, 1, lightpath_index
            ),
        ),
        lambda x: x,
        state
    )
    capacity_mask = jnp.where(state.link_capacity_array <= 0., -1., 0.)
    over_capacity_mask = jnp.where(state.link_capacity_array < 0., -1., 0.)
    # Undo link_capacity_update if over capacity
    # N.B. this will fail if requested capacity is greater than total original capacity of lightpath
    lightpath_capacity_before_action = jax.lax.cond(
        lightpath_existing_check,
        lambda x: curr_lightpath_capacity,  # Subtract requested_datarate from current lightpath
        lambda x: 1e6,  # Empty slots have high capacity (1e6)
        # Get initial capacity of lightpath - request
        None,
    )
    state = state.replace(
        link_capacity_array=jnp.where(over_capacity_mask == -1, lightpath_capacity_before_action, state.link_capacity_array)
    )
    # Total mask will be 0 if space still available, -1 if capacity is zero or -2 if over capacity
    total_mask = capacity_mask + over_capacity_mask
    # Update link_slot_array and link_slot_departure_array
    state = state.replace(
        link_slot_array=total_mask,
        link_slot_departure_array=vmap_update_path_links(state.link_slot_departure_array, path,
                                                                 initial_slot_index, 1,
                                                                 state.current_time + state.holding_time)
    )
    return state

implement_node_action(state, s_node, d_node, s_request, d_request, n=2)

Update node capacity, node resource and node departure arrays

Parameters:

Name Type Description Default
state State

current state

 required
s_node int

source node

 required
d_node int

destination node

 required
s_request int

source node request

 required
d_request int

destination node request

 required
n int

number of nodes to implement. Defaults to 2.

 2

Returns:

Name Type Description
State EnvState

updated state
Source code in xlron/environments/env_funcs.py 
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
def implement_node_action(state: EnvState, s_node: chex.Array, d_node: chex.Array, s_request: chex.Array, d_request: chex.Array, n=2) -> EnvState:
    """Update node capacity, node resource and node departure arrays

    Args:
        state (State): current state
        s_node (int): source node
        d_node (int): destination node
        s_request (int): source node request
        d_request (int): destination node request
        n (int, optional): number of nodes to implement. Defaults to 2.

    Returns:
        State: updated state
    """
    node_indices = jnp.arange(state.node_capacity_array.shape[0])

    curr_selected_nodes = jnp.zeros(state.node_capacity_array.shape[0])
    # d_request -ve so that selected node is +ve (so that argmin works correctly for node resource array update)
    # curr_selected_nodes is N x 1 array, with requested node resources at index of selected node
    curr_selected_nodes = update_node_array(node_indices, curr_selected_nodes, d_node, -d_request)
    curr_selected_nodes = jax.lax.cond(n == 2, lambda x: update_node_array(*x), lambda x: x[1], (node_indices, curr_selected_nodes, s_node, -s_request))

    node_capacity_array = state.node_capacity_array - curr_selected_nodes

    node_resource_array = vmap_update_node_resources(state.node_resource_array, curr_selected_nodes)

    node_departure_array = vmap_update_node_departure(state.node_departure_array, curr_selected_nodes, -state.current_time-state.holding_time)

    state = state.replace(
        node_capacity_array=node_capacity_array,
        node_resource_array=node_resource_array,
        node_departure_array=node_departure_array
    )

    return state

implement_path_action(state, path, initial_slot_index, num_slots)

Update link-slot and link-slot departure arrays. Times are set to negative until turned positive by finalisation (after checks).

Parameters:

Name Type Description Default
state State

current state

 required
path int

path to implement

 required
initial_slot_index int

initial slot index

 required
num_slots int

number of slots to implement

 required
Source code in xlron/environments/env_funcs.py 
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
def implement_path_action(state: EnvState, path: chex.Array, initial_slot_index: chex.Array, num_slots: chex.Array) -> EnvState:
    """Update link-slot and link-slot departure arrays.
    Times are set to negative until turned positive by finalisation (after checks).

    Args:
        state (State): current state
        path (int): path to implement
        initial_slot_index (int): initial slot index
        num_slots (int): number of slots to implement
    """
    state = state.replace(
        link_slot_array=vmap_update_path_links(state.link_slot_array, path, initial_slot_index, num_slots, 1),
        link_slot_departure_array=vmap_update_path_links(state.link_slot_departure_array, path, initial_slot_index, num_slots, state.current_time+state.holding_time)
    )
    return state

implement_vone_action(state, action, total_actions, remaining_actions, params)

Implement action to assign nodes (1, 2, or 0 nodes assigned per action) and assign slots and links for lightpath.

Parameters:

Name Type Description Default
state EnvState

current state

 required
action Array

action to implement (node, node, path_slot_action)

 required
total_actions Scalar

total number of actions to implement for current request

 required
remaining_actions Scalar

remaining actions to implement

 required
k

number of paths to consider

 required
N

number of nodes to assign

 required

Returns:

Name Type Description
state

updated state
Source code in xlron/environments/env_funcs.py 
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
@partial(jax.jit, static_argnums=(4,))
def implement_vone_action(
        state: EnvState,
        action: chex.Array,
        total_actions: chex.Scalar,
        remaining_actions: chex.Scalar,
        params: EnvParams,
):
    """Implement action to assign nodes (1, 2, or 0 nodes assigned per action) and assign slots and links for lightpath.

    Args:
        state: current state
        action: action to implement (node, node, path_slot_action)
        total_actions: total number of actions to implement for current request
        remaining_actions: remaining actions to implement
        k: number of paths to consider
        N: number of nodes to assign

    Returns:
        state: updated state
    """
    request = jax.lax.dynamic_slice(state.request_array[0], ((remaining_actions-1)*2, ), (3, ))
    node_request_s = jax.lax.dynamic_slice(request, (2, ), (1, ))
    requested_datarate = jax.lax.dynamic_slice(request, (1,), (1,))
    node_request_d = jax.lax.dynamic_slice(request, (0, ), (1, ))
    nodes = action[::2]
    path_index, initial_slot_index = process_path_action(state, params, action[1])
    path = get_paths(params, nodes)[path_index]
    se = get_paths_se(params, nodes)[path_index] if params.consider_modulation_format else jnp.array([1])
    num_slots = required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband)

    # jax.debug.print("state.request_array {}", state.request_array, ordered=True)
    # jax.debug.print("path {}", path, ordered=True)
    # jax.debug.print("slots {}", jnp.max(jnp.where(path.reshape(-1,1) == 1, state.link_slot_array, jnp.zeros(params.num_links).reshape(-1,1)), axis=0), ordered=True)
    # jax.debug.print("path_index {}", path_index, ordered=True)
    # jax.debug.print("initial_slot_index {}", initial_slot_index, ordered=True)
    # jax.debug.print("requested_datarate {}", requested_datarate, ordered=True)
    # jax.debug.print("request {}", request, ordered=True)
    # jax.debug.print("se {}", se, ordered=True)
    # jax.debug.print("num_slots {}", num_slots, ordered=True)

    n_nodes = jax.lax.cond(
        total_actions == remaining_actions,
        lambda x: 2, lambda x: 1,
        (total_actions, remaining_actions))
    path_action_only_check = path_action_only(state.request_array[1], state.action_counter, remaining_actions)

    state = jax.lax.cond(
        path_action_only_check,
        lambda x: x[0],
        lambda x: implement_node_action(x[0], x[1], x[2], x[3], x[4], n=x[5]),
        (state, nodes[0], nodes[1], node_request_s, node_request_d, n_nodes)
    )

    state = implement_path_action(state, path, initial_slot_index, num_slots)

    return state

init_action_counter()

Initialize action counter. First index is num unique nodes, second index is total steps, final is remaining steps until completion of request. Only used in VONE environments.

Source code in xlron/environments/env_funcs.py 
539
540
541
542
543
544
def init_action_counter():
    """Initialize action counter.
    First index is num unique nodes, second index is total steps, final is remaining steps until completion of request.
    Only used in VONE environments.
    """
    return jnp.zeros(3, dtype=jnp.int32)

init_action_history(params)

Initialize action history

Source code in xlron/environments/env_funcs.py 
574
575
576
577
@partial(jax.jit, static_argnums=(0,))
def init_action_history(params: EnvParams):
    """Initialize action history"""
    return jnp.full(params.max_edges*2+1, -1)

init_active_lightpaths_array(params)

Initialise active lightpath array. Stores path indices of all active paths on the network in a 1 x M array. M is MIN(max_requests, num_links * link_resources / min_slots). min_slots is the minimum number of slots required for a lightpath i.e. max(values_bw)/ slot_size.

Parameters:

Name Type Description Default
params RSAGNModelEnvParams

Environment parameters

 required

Returns: jnp.array: Active path array (default value -1, empty path)

Source code in xlron/environments/env_funcs.py 
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
def init_active_lightpaths_array(params: RSAGNModelEnvParams):
    """Initialise active lightpath array. Stores path indices of all active paths on the network in a 1 x M array.
    M is MIN(max_requests, num_links * link_resources / min_slots).
    min_slots is the minimum number of slots required for a lightpath i.e. max(values_bw)/ slot_size.

    Args:
        params (RSAGNModelEnvParams): Environment parameters
    Returns:
        jnp.array: Active path array (default value -1, empty path)
    """
    total_slots = params.num_links * params.link_resources  # total slots on networks
    min_slots = jnp.max(params.values_bw.val) / params.slot_size  # minimum number of slots required for lightpath
    return jnp.full((min(int(params.max_requests), int(total_slots / min_slots)),), -1)

init_active_lightpaths_array_departure(params)

Initialise active lightpath array. Stores path indices of all active paths on the network in a 1 x M array. M is MIN(max_requests, num_links * link_resources / min_slots). min_slots is the minimum number of slots required for a lightpath i.e. max(values_bw)/ slot_size.

Parameters:

Name Type Description Default
params RSAGNModelEnvParams

Environment parameters

 required

Returns: jnp.array: Active path array (default value -1, empty path)

Source code in xlron/environments/env_funcs.py 
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
def init_active_lightpaths_array_departure(params: RSAGNModelEnvParams):
    """Initialise active lightpath array. Stores path indices of all active paths on the network in a 1 x M array.
    M is MIN(max_requests, num_links * link_resources / min_slots).
    min_slots is the minimum number of slots required for a lightpath i.e. max(values_bw)/ slot_size.

    Args:
        params (RSAGNModelEnvParams): Environment parameters
    Returns:
        jnp.array: Active path array (default value -1, empty path)
    """
    total_slots = params.num_links * params.link_resources  # total slots on networks
    min_slots = jnp.max(params.values_bw.val) / params.slot_size  # minimum number of slots required for lightpath
    return jnp.full((min(int(params.max_requests), int(total_slots / min_slots)),), 0.)

init_active_path_array(params)

Initialise active path array. Stores details of full path utilised by lightpath on each frequency slot. Args: params (EnvParams): Environment parameters Returns: jnp.array: Active path array

Source code in xlron/environments/env_funcs.py 
2569
2570
2571
2572
2573
2574
2575
2576
def init_active_path_array(params: EnvParams):
    """Initialise active path array. Stores details of full path utilised by lightpath on each frequency slot.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Active path array
    """
    return jnp.full((params.num_links, params.link_resources, params.num_links), 0.)

init_channel_centre_bw_array(params)

Initialise channel centre array. Args: params (EnvParams): Environment parameters Returns: jnp.array: Channel centre array

Source code in xlron/environments/env_funcs.py 
2549
2550
2551
2552
2553
2554
2555
2556
def init_channel_centre_bw_array(params: EnvParams):
    """Initialise channel centre array.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Channel centre array
    """
    return jnp.full((params.num_links, params.link_resources), 0.)

init_channel_power_array(params)

Initialise channel power array.

Parameters:

Name Type Description Default
params EnvParams

Environment parameters

 required

Returns: jnp.array: Channel power array

Source code in xlron/environments/env_funcs.py 
2538
2539
2540
2541
2542
2543
2544
2545
2546
def init_channel_power_array(params: EnvParams):
    """Initialise channel power array.

    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Channel power array
    """
    return jnp.full((params.num_links, params.link_resources), 0.)

init_graph_tuple(state, params, adj, exclude_source_dest=False)

Initialise graph tuple for use with Jraph GNNs. Args: state (EnvState): Environment state params (EnvParams): Environment parameters adj (jnp.array): Adjacency matrix of the graph Returns: jraph.GraphsTuple: Graph tuple

Source code in xlron/environments/env_funcs.py 
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
@partial(jax.jit, static_argnums=(1, 3))
def init_graph_tuple(state: EnvState, params: EnvParams, adj: jnp.array, exclude_source_dest: bool=False) -> jraph.GraphsTuple:
    """Initialise graph tuple for use with Jraph GNNs.
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
        adj (jnp.array): Adjacency matrix of the graph
    Returns:
        jraph.GraphsTuple: Graph tuple
    """
    senders = params.edges.val.T[0]
    receivers = params.edges.val.T[1]

    if exclude_source_dest:
        source_dest_features = jnp.zeros((params.num_nodes, 2))
    else:
        # Get source and dest from request array
        source_dest, _ = read_rsa_request(state.request_array)
        source, dest = source_dest[0], source_dest[2]
        # One-hot encode source and destination (2 additional features)
        source_dest_features = jnp.zeros((params.num_nodes, 2))
        source_dest_features = source_dest_features.at[source.astype(jnp.int32), 0].set(1)
        source_dest_features = source_dest_features.at[dest.astype(jnp.int32), 1].set(-1)

    spectral_features = get_spectral_features(adj, num_features=3)

    if params.__class__.__name__ in ["RSAGNModelEnvParams", "RMSAGNModelEnvParams"]:
        # Normalize by max parameters (converted to linear units)
        max_power = isrs_gn_model.from_dbm(params.max_power)
        normalized_power = jnp.round(state.channel_power_array / max_power, 3)
        max_snr = isrs_gn_model.from_db(params.max_snr)
        normalized_snr = jnp.round(state.link_snr_array / max_snr, 3)
        edge_features = jnp.stack([normalized_snr, normalized_power], axis=-1)
        node_features = jnp.concatenate([spectral_features, source_dest_features], axis=-1)
    elif params.__class__.__name__ == "VONEEnvParams":
        edge_features = state.link_slot_array  # [n_edges] or [n_edges, ...]
        node_features = getattr(state, "node_capacity_array", jnp.zeros(params.num_nodes))
        node_features = node_features.reshape(-1, 1)
        node_features = jnp.concatenate([node_features, spectral_features, source_dest_features], axis=-1)
    else:
        edge_features = state.link_slot_array  # [n_edges] or [n_edges, ...]
        node_features = jnp.concatenate([spectral_features, source_dest_features], axis=-1)

    if params.disable_node_features:
        node_features = jnp.zeros((1,))

    # Handle undirected graphs (duplicate edges after normalization)
    if not params.directed_graph:
        senders_ = jnp.concatenate([senders, receivers])
        receivers = jnp.concatenate([receivers, senders])
        senders = senders_
        edge_features = jnp.repeat(edge_features, 2, axis=0)

    return jraph.GraphsTuple(
        nodes=node_features,
        edges=edge_features,
        senders=senders,
        receivers=receivers,
        n_node=jnp.reshape(params.num_nodes, (1,)),
        n_edge=jnp.reshape(len(senders), (1,)),
        globals=jnp.reshape(state.request_array, (1, -1)),
    )

Initialise link capacity array. Represents available data rate for lightpath on each link. Default is high value (1e6) for unoccupied slots. Once lightpath established, capacity is determined by corresponding entry in path capacity array.

Source code in xlron/environments/env_funcs.py 
2170
2171
2172
2173
2174
def init_link_capacity_array(params):
    """Initialise link capacity array. Represents available data rate for lightpath on each link.
    Default is high value (1e6) for unoccupied slots. Once lightpath established, capacity is determined by
    corresponding entry in path capacity array."""
    return jnp.full((params.num_links, params.link_resources), 1e6)

Initialise link length array. Args: graph (nx.Graph): NetworkX graph Returns:

Source code in xlron/environments/env_funcs.py 
185
186
187
188
189
190
191
192
193
194
195
def init_link_length_array(graph: nx.Graph) -> chex.Array:
    """Initialise link length array.
    Args:
        graph (nx.Graph): NetworkX graph
    Returns:

    """
    link_lengths = []
    for edge in sorted(graph.edges):
        link_lengths.append(graph.edges[edge]["weight"])
    return jnp.array(link_lengths)

Initialise link length array for environements that use GN model of physical layer. We assume each link has spans of equal length.

Parameters:

Name Type Description Default
graph Graph

NetworkX graph

 required

Returns: jnp.array: Link length array (L x max_spans)

Source code in xlron/environments/env_funcs.py 
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
def init_link_length_array_gn_model(graph: nx.Graph, max_span_length: int,  max_spans: int) -> chex.Array:
    """Initialise link length array for environements that use GN model of physical layer.
    We assume each link has spans of equal length.

    Args:
        graph (nx.Graph): NetworkX graph
    Returns:
        jnp.array: Link length array (L x max_spans)
    """
    link_lengths = []
    directed = graph.is_directed()
    graph = graph.to_undirected()
    edges = sorted(graph.edges)
    for edge in edges:
        link_lengths.append(graph.edges[edge]["weight"])
    if directed:
        for edge in edges:
            link_lengths.append(graph.edges[edge]["weight"])
    span_length_array = []
    for length in link_lengths:
        num_spans = math.ceil(length / max_span_length)
        avg_span_length = length / num_spans
        span_lengths = [avg_span_length] * num_spans
        span_lengths.extend([0] * (max_spans - num_spans))
        span_length_array.append(span_lengths)
    return jnp.array(span_length_array)

Initialize empty (all zeroes) link-slot array. 0 means slot is free, -1 means occupied. Args: params (EnvParams): Environment parameters Returns: jnp.array: Link slot array (E x S) where E is number of edges and S is number of slots

Source code in xlron/environments/env_funcs.py 
500
501
502
503
504
505
506
507
@partial(jax.jit, static_argnums=(0,))
def init_link_slot_array(params: EnvParams):
    """Initialize empty (all zeroes) link-slot array. 0 means slot is free, -1 means occupied.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Link slot array (E x S) where E is number of edges and S is number of slots"""
    return jnp.zeros((params.num_links, params.link_resources))

Initialize link mask

Source code in xlron/environments/env_funcs.py 
527
528
529
530
@partial(jax.jit, static_argnums=(0, 1))
def init_link_slot_mask(params: EnvParams, agg: int = 1):
    """Initialize link mask"""
    return jnp.ones(params.k_paths*math.ceil(params.link_resources / agg))

Initialise signal-to-noise ratio (SNR) array. Args: params (EnvParams): Environment parameters Returns: jnp.array: SNR array

Source code in xlron/environments/env_funcs.py 
2527
2528
2529
2530
2531
2532
2533
2534
2535
def init_link_snr_array(params: EnvParams):
    """Initialise signal-to-noise ratio (SNR) array.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: SNR array
    """
    # The SNR is kept in linear units to allow summation of 1/SNR across links
    return jnp.full((params.num_links, params.link_resources), -1e5)

init_mod_format_mask(params)

Initialize link mask

Source code in xlron/environments/env_funcs.py 
533
534
535
536
@partial(jax.jit, static_argnums=(0,))
def init_mod_format_mask(params: EnvParams):
    """Initialize link mask"""
    return jnp.full((params.k_paths*params.link_resources,), -1.0)

init_modulation_format_index_array(params)

Initialise modulation format index array. Args: params (EnvParams): Environment parameters Returns: jnp.array: Modulation format index array

Source code in xlron/environments/env_funcs.py 
2559
2560
2561
2562
2563
2564
2565
2566
def init_modulation_format_index_array(params: EnvParams):
    """Initialise modulation format index array.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Modulation format index array
    """
    return jnp.full((params.num_links, params.link_resources), -1)  # -1 so that highest order is assumed (closest to Gaussian)

init_modulations_array(modulations_filepath=None)

Initialise array of maximum spectral efficiency for modulation format on path.

Parameters:

Name Type Description Default
modulations_filepath str

Path to CSV file containing modulation formats. Defaults to None.

 None

Returns: jnp.array: Array of maximum spectral efficiency for modulation format on path. First two columns are maximum path length and spectral efficiency.

Source code in xlron/environments/env_funcs.py 
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
def init_modulations_array(modulations_filepath: str = None):
    """Initialise array of maximum spectral efficiency for modulation format on path.

    Args:
        modulations_filepath (str, optional): Path to CSV file containing modulation formats. Defaults to None.
    Returns:
        jnp.array: Array of maximum spectral efficiency for modulation format on path.
        First two columns are maximum path length and spectral efficiency.
    """
    f = pathlib.Path(modulations_filepath) if modulations_filepath else (
            pathlib.Path(__file__).parents[1].absolute() / "data" / "modulations" / "modulations.csv")
    modulations = np.genfromtxt(f, delimiter=',')
    # Drop empty first row (headers) and column (name)
    modulations = modulations[1:, 1:]
    return jnp.array(modulations)

init_node_capacity_array(params)

Initialize node array with uniform resources. Args: params (EnvParams): Environment parameters Returns: jnp.array: Node capacity array (N x 1) where N is number of nodes

Source code in xlron/environments/env_funcs.py 
490
491
492
493
494
495
496
497
@partial(jax.jit, static_argnums=(0,))
def init_node_capacity_array(params: EnvParams):
    """Initialize node array with uniform resources.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Node capacity array (N x 1) where N is number of nodes"""
    return jnp.array([params.node_resources] * params.num_nodes, dtype=jnp.float32)

init_node_mask(params)

Initialize node mask

Source code in xlron/environments/env_funcs.py 
521
522
523
524
@partial(jax.jit, static_argnums=(0,))
def init_node_mask(params: EnvParams):
    """Initialize node mask"""
    return jnp.ones(params.num_nodes)

init_node_resource_array(params)

Array to track node resources occupied by virtual nodes

Source code in xlron/environments/env_funcs.py 
568
569
570
571
@partial(jax.jit, static_argnums=(0,))
def init_node_resource_array(params: EnvParams):
    """Array to track node resources occupied by virtual nodes"""
    return jnp.zeros((params.num_nodes, params.node_resources), dtype=jnp.float32)

init_path_capacity_array(link_length_array, path_link_array, min_request=1, scale_factor=1.0, alpha=0.0002, NF=4.5, B=10000000000000.0, R_s=100000000000.0, beta_2=-2.17e-26, gamma=0.0012, L_s=100000.0, lambda0=1.55e-06)

Calculated from Nevin paper: https://api.repository.cam.ac.uk/server/api/core/bitstreams/b80e7a9c-a86b-4b30-a6d6-05017c60b0c8/content

Parameters:

Name Type Description Default
link_length_array Array

Array of link lengths

 required
path_link_array Array

Array of links on paths

 required
min_request int

Minimum data rate request size. Defaults to 100 GBps.

 1
scale_factor float

Scale factor for link capacity. Defaults to 1.0.

 1.0
alpha float

Fibre attenuation coefficient. Defaults to 0.2e-3 /m

 0.0002
NF float

Amplifier noise figure. Defaults to 4.5 dB.

 4.5
B float

Total modulated bandwidth. Defaults to 10e12 Hz.

 10000000000000.0
R_s float

Symbol rate. Defaults to 100e9 Baud.

 100000000000.0
beta_2 float

Dispersion parameter. Defaults to -21.7e-27 s^2/m.

 -2.17e-26
gamma float

Nonlinear coefficient. Defaults to 1.2e-3 /W/m.

 0.0012
L_s float

Span length. Defaults to 100e3 m.

 100000.0
lambda0 float

Wavelength. Defaults to 1550e-9 m.

 1.55e-06

Returns:

Type Description
Array

chex.Array: Array of link capacities in Gbps
Source code in xlron/environments/env_funcs.py 
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
def init_path_capacity_array(
        link_length_array: chex.Array,
        path_link_array: chex.Array,
        min_request=1,  # Minimum data rate request size
        scale_factor=1.0,  # Scale factor for link capacity
        alpha=0.2e-3,  # Fibre attenuation coefficient
        NF=4.5,  # Amplifier noise figure
        B=10e12,  # Total modulated bandwidth
        R_s=100e9,  # Symbol rate
        beta_2=-21.7e-27,  # Dispersion parameter
        gamma=1.2e-3,  # Nonlinear coefficient
        L_s=100e3,  # span length
        lambda0=1550e-9,  # Wavelength
) -> chex.Array:
    """Calculated from Nevin paper:
    https://api.repository.cam.ac.uk/server/api/core/bitstreams/b80e7a9c-a86b-4b30-a6d6-05017c60b0c8/content

    Args:
        link_length_array (chex.Array): Array of link lengths
        path_link_array (chex.Array): Array of links on paths
        min_request (int, optional): Minimum data rate request size. Defaults to 100 GBps.
        scale_factor (float, optional): Scale factor for link capacity. Defaults to 1.0.
        alpha (float, optional): Fibre attenuation coefficient. Defaults to 0.2e-3 /m
        NF (float, optional): Amplifier noise figure. Defaults to 4.5 dB.
        B (float, optional): Total modulated bandwidth. Defaults to 10e12 Hz.
        R_s (float, optional): Symbol rate. Defaults to 100e9 Baud.
        beta_2 (float, optional): Dispersion parameter. Defaults to -21.7e-27 s^2/m.
        gamma (float, optional): Nonlinear coefficient. Defaults to 1.2e-3 /W/m.
        L_s (float, optional): Span length. Defaults to 100e3 m.
        lambda0 (float, optional): Wavelength. Defaults to 1550e-9 m.

    Returns:
        chex.Array: Array of link capacities in Gbps
    """
    path_length_array = jnp.dot(path_link_array, link_length_array)
    path_capacity_array = calculate_path_capacity(
        path_length_array,
        min_request=min_request,
        scale_factor=scale_factor,
        alpha=alpha,
        NF=NF,
        B=B,
        R_s=R_s,
        beta_2=beta_2,
        gamma=gamma,
        L_s=L_s,
        lambda0=lambda0,
    )
    return path_capacity_array

init_path_index_array(params)

Initialise path index array. Represents index of lightpath occupying each slot.

Source code in xlron/environments/env_funcs.py 
2177
2178
2179
def init_path_index_array(params):
    """Initialise path index array. Represents index of lightpath occupying each slot."""
    return jnp.full((params.num_links, params.link_resources), -1)

init_path_length_array(path_link_array, graph)

Initialise path length array.

Parameters:

Name Type Description Default
path_link_array Array

Path-link array

 required
graph Graph

NetworkX graph

 required

Returns: chex.Array: Path length array

Source code in xlron/environments/env_funcs.py 
331
332
333
334
335
336
337
338
339
340
341
342
def init_path_length_array(path_link_array: chex.Array, graph: nx.Graph) -> chex.Array:
    """Initialise path length array.

    Args:
        path_link_array (chex.Array): Path-link array
        graph (nx.Graph): NetworkX graph
    Returns:
        chex.Array: Path length array
    """
    link_length_array = init_link_length_array(graph)
    path_lengths = jnp.dot(path_link_array, link_length_array)
    return path_lengths

Initialise path-link array. Each path is defined by a link utilisation array (one row in the path-link array). 1 indicates link corresponding to index is used, 0 indicates not used.

Parameters:

Name Type Description Default
graph Graph

NetworkX graph

 required
k int

Number of paths

 required
disjoint bool

Whether to use edge-disjoint paths. Defaults to False.

 False
weight str

Sort paths by edge attribute. Defaults to "weight".

 'weight'
directed bool

Whether graph is directed. Defaults to False.

 False
modulations_array Array

Array of maximum spectral efficiency for modulation format on path. Defaults to None.

 None
rwa_lr bool

Whether the environment is RWA with lightpath reuse (affects path ordering).

 False
path_snr bool

If GN model is used, include extra row of zeroes for unutilised paths

 False

Returns:

Type Description
Array

chex.Array: Path-link array (N(N-1)*k x E) where N is number of nodes, E is number of edges, k is number of shortest paths
Source code in xlron/environments/env_funcs.py 
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
def init_path_link_array(
        graph: nx.Graph,
        k: int,
        disjoint: bool = False,
        weight: str = "weight",
        directed: bool = False,
        modulations_array: chex.Array = None,
        rwa_lr: bool = False,
        scale_factor: float = 1.0,
        path_snr: bool = False,
) -> chex.Array:
    """Initialise path-link array.
    Each path is defined by a link utilisation array (one row in the path-link array).
    1 indicates link corresponding to index is used, 0 indicates not used.

    Args:
        graph (nx.Graph): NetworkX graph
        k (int): Number of paths
        disjoint (bool, optional): Whether to use edge-disjoint paths. Defaults to False.
        weight (str, optional): Sort paths by edge attribute. Defaults to "weight".
        directed (bool, optional): Whether graph is directed. Defaults to False.
        modulations_array (chex.Array, optional): Array of maximum spectral efficiency for modulation format on path. Defaults to None.
        rwa_lr (bool, optional): Whether the environment is RWA with lightpath reuse (affects path ordering).
        path_snr (bool, optional): If GN model is used, include extra row of zeroes for unutilised paths
        to ensure correct SNR calculation for empty paths (path index -1).

    Returns:
        chex.Array: Path-link array (N(N-1)*k x E) where N is number of nodes, E is number of edges, k is number of shortest paths
    """
    def get_k_shortest_paths(g, source, target, k, weight):
        return list(
            islice(nx.shortest_simple_paths(g, source, target, weight=weight), k)
        )

    def get_k_disjoint_shortest_paths(g, source, target, k, weight):
        k_paths_disjoint_unsorted = list(nx.edge_disjoint_paths(g, source, target))
        k_paths_shortest = get_k_shortest_paths(g, source, target, k, weight=weight)

        # Keep disjoint paths and add unique shortest paths until k paths reached
        disjoint_ids = [tuple(path) for path in k_paths_disjoint_unsorted]
        k_paths = k_paths_disjoint_unsorted
        for path in k_paths_shortest:
            if tuple(path) not in disjoint_ids:
                k_paths.append(path)
        k_paths = k_paths[:k]
        return k_paths

    paths = []
    edges = sorted(graph.edges)

    # Get the k-shortest paths for each node pair
    k_path_collections = []
    get_paths = get_k_disjoint_shortest_paths if disjoint else get_k_shortest_paths
    for node_pair in combinations(graph.nodes, 2):

        k_paths = get_paths(graph, node_pair[0], node_pair[1], k, weight=weight)
        k_path_collections.append(k_paths)

    if directed:  # Get paths in reverse direction
        for node_pair in combinations(graph.nodes, 2):
            k_paths_rev = get_paths(graph, node_pair[1], node_pair[0], k, weight=weight)
            k_path_collections.append(k_paths_rev)

    # Sort the paths for each node pair
    max_missing_paths = 0
    for k_paths in k_path_collections:

        source, dest = k_paths[0][0], k_paths[0][-1]

        # Sort the paths by # of hops then by length, or just length
        path_lengths = [nx.path_weight(graph, path, weight='weight') for path in k_paths]
        path_num_links = [len(path) - 1 for path in k_paths]

        # Get maximum spectral efficiency for modulation format on path
        if modulations_array is not None and rwa_lr is not True:
            se_of_path = []
            modulations_array = modulations_array[::-1]
            for length in path_lengths:
                for modulation in modulations_array:
                    if length <= modulation[0]:
                        se_of_path.append(modulation[1])
                        break
            # Sorting by the num_links/se instead of just path length is observed to improve performance
            path_weighting = [num_links/se for se, num_links in zip(se_of_path, path_num_links)]
        elif rwa_lr:
            path_capacity = [float(calculate_path_capacity(path_length, scale_factor=scale_factor)) for path_length in path_lengths]
            path_weighting = [num_links/path_capacity for num_links, path_capacity in zip(path_num_links, path_capacity)]
        elif weight is None:
            path_weighting = path_num_links
        else:
            path_weighting = path_lengths

        # if less then k unique paths, add empty paths
        empty_path = [0] * len(graph.edges)
        num_missing_paths = k - len(k_paths)
        max_missing_paths = max(max_missing_paths, num_missing_paths)
        k_paths = k_paths + [empty_path] * num_missing_paths
        path_weighting = path_weighting + [1e6] * num_missing_paths
        path_lengths = path_lengths + [1e6] * num_missing_paths

        # Sort by number of links then by length (or just by length if weight is specified)
        unsorted_paths = zip(k_paths, path_weighting, path_lengths)
        k_paths_sorted = [(source, dest, weighting, path) for path, weighting, _ in sorted(unsorted_paths, key=lambda x: (x[1], 1/x[2]) if weight is None else x[2])]

        # Keep only first k paths
        k_paths_sorted = k_paths_sorted[:k]

        prev_link_usage = empty_path
        for k_path in k_paths_sorted:
            k_path = k_path[-1]
            link_usage = [0]*len(graph.edges)  # Initialise empty path
            if sum(k_path) == 0:
                link_usage = prev_link_usage
            else:
                for i in range(len(k_path)-1):
                    s, d = k_path[i], k_path[i + 1]
                    for edge_index, edge in enumerate(edges):
                        condition = (edge[0] == s and edge[1] == d) if directed else \
                            ((edge[0] == s and edge[1] == d) or (edge[0] == d and edge[1] == s))
                        if condition:
                            link_usage[edge_index] = 1
            path = link_usage
            prev_link_usage = link_usage
            paths.append(path)

    # If using GN model, add extra row of zeroes for empty paths for SNR calculation
    if path_snr:
        empty_path = [0] * len(graph.edges)
        paths.append(empty_path)

    return jnp.array(paths, dtype=jnp.float32)

init_path_se_array(path_length_array, modulations_array)

Initialise array of maximum spectral efficiency for highest-order modulation format on path.

Parameters:

Name Type Description Default
path_length_array array

Array of path lengths

 required
modulations_array array

Array of maximum spectral efficiency for modulation format on path

 required

Returns:

Type Description

jnp.array: Array of maximum spectral efficiency for on path
Source code in xlron/environments/env_funcs.py 
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
def init_path_se_array(path_length_array, modulations_array):
    """Initialise array of maximum spectral efficiency for highest-order modulation format on path.

    Args:
        path_length_array (jnp.array): Array of path lengths
        modulations_array (jnp.array): Array of maximum spectral efficiency for modulation format on path

    Returns:
        jnp.array: Array of maximum spectral efficiency for on path
    """
    se_list = []
    # Flip the modulation array so that the shortest path length is first
    modulations_array = modulations_array[::-1]
    for length in path_length_array:
        for modulation in modulations_array:
            if length <= modulation[0]:
                se_list.append(modulation[1])
                break
    return jnp.array(se_list)

init_rsa_request_array()

Initialize request array

Source code in xlron/environments/env_funcs.py 
516
517
518
def init_rsa_request_array():
    """Initialize request array"""
    return jnp.zeros(3)

init_traffic_matrix(key, params)

Initialize traffic matrix. Allows for random traffic matrix or uniform traffic matrix. Source-dest traffic requests are sampled probabilistically from the resulting traffic matrix.

Parameters:

Name Type Description Default
key PRNGKey

PRNG key

 required
params EnvParams

Environment parameters

 required

Returns:

Type Description

jnp.array: Traffic matrix
Source code in xlron/environments/env_funcs.py 
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
@partial(jax.jit, static_argnums=(1,))
def init_traffic_matrix(key: chex.PRNGKey, params: EnvParams):
    """Initialize traffic matrix. Allows for random traffic matrix or uniform traffic matrix.
    Source-dest traffic requests are sampled probabilistically from the resulting traffic matrix.

    Args:
        key (chex.PRNGKey): PRNG key
        params (EnvParams): Environment parameters

    Returns:
        jnp.array: Traffic matrix
    """
    if params.random_traffic:
        traffic_matrix = jax.random.uniform(key, shape=(params.num_nodes, params.num_nodes))
    else:
        traffic_matrix = jnp.ones((params.num_nodes, params.num_nodes))
    diag_elements = jnp.diag_indices_from(traffic_matrix)
    # Set main diagonal to zero so no requests from node to itself
    traffic_matrix = traffic_matrix.at[diag_elements].set(0)
    traffic_matrix = normalise_traffic_matrix(traffic_matrix)
    return traffic_matrix

init_virtual_topology_patterns(pattern_names)

Initialise virtual topology patterns. First 3 digits comprise the "action counter": first index is num unique nodes, second index is total steps, final is remaining steps until completion of request. Remaining digits define the topology pattern, with 1 to indicate links and other positive integers are node indices.

Parameters:

Name Type Description Default
pattern_names list

List of virtual topology pattern names

 required

Returns:

Type Description
Array

chex.Array: Array of virtual topology patterns
Source code in xlron/environments/env_funcs.py 
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
def init_virtual_topology_patterns(pattern_names: str) -> chex.Array:
    """Initialise virtual topology patterns.
    First 3 digits comprise the "action counter": first index is num unique nodes, second index is total steps,
    final is remaining steps until completion of request.
    Remaining digits define the topology pattern, with 1 to indicate links and other positive integers are node indices.

    Args:
        pattern_names (list): List of virtual topology pattern names

    Returns:
        chex.Array: Array of virtual topology patterns
    """
    patterns = []
    # TODO - Allow 2 node requests in VONE (check if any modifications necessary other than below)
    #if "2_bus" in pattern_names:
    #    patterns.append([2, 1, 1, 2, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0])
    if "3_bus" in pattern_names:
        patterns.append([3, 2, 2, 2, 1, 3, 1, 4])
    if "3_ring" in pattern_names:
        patterns.append([3, 3, 3, 2, 1, 3, 1, 4, 1, 2])
    if "4_bus" in pattern_names:
        patterns.append([4, 3, 3, 2, 1, 3, 1, 4, 1, 5])
    if "4_ring" in pattern_names:
        patterns.append([4, 4, 4, 2, 1, 3, 1, 4, 1, 5, 1, 2])
    if "5_bus" in pattern_names:
        patterns.append([5, 4, 4, 2, 1, 3, 1, 4, 1, 5, 1, 6])
    if "5_ring" in pattern_names:
        patterns.append([5, 5, 5, 2, 1, 3, 1, 4, 1, 5, 1, 6, 1, 2])
    if "6_bus" in pattern_names:
        patterns.append([6, 5, 5, 2, 1, 3, 1, 4, 1, 5, 1, 6, 1, 7])
    max_length = max([len(pattern) for pattern in patterns])
    # Pad patterns with zeroes to match longest
    for pattern in patterns:
        pattern.extend([0]*(max_length-len(pattern)))
    return jnp.array(patterns, dtype=jnp.int32)

init_vone_request_array(params)

Initialize request array either with uniform resources

Source code in xlron/environments/env_funcs.py 
510
511
512
513
@partial(jax.jit, static_argnums=(0,))
def init_vone_request_array(params: EnvParams):
    """Initialize request array either with uniform resources"""
    return jnp.zeros((2, params.max_edges*2+1, ))

make_graph(topology_name='conus', topology_directory=None)

Create graph from topology definition. Topologies must be defined in JSON format in the topologies directory and named as the topology name with .json extension.

Parameters:

Name Type Description Default
topology_name str

topology name

 'conus'
topology_directory str

topology directory

 None

Returns:

Name Type Description
graph

graph
Source code in xlron/environments/env_funcs.py 
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
def make_graph(topology_name: str = "conus", topology_directory: str = None):
    """Create graph from topology definition.
    Topologies must be defined in JSON format in the topologies directory and
    named as the topology name with .json extension.

    Args:
        topology_name: topology name
        topology_directory: topology directory

    Returns:
        graph: graph
    """
    topology_path = pathlib.Path(topology_directory) if topology_directory else (
            pathlib.Path(__file__).parents[1].absolute() / "data" / "topologies")
    # Create topology
    if topology_name == "4node":
        # 4 node ring
        graph = nx.from_numpy_array(np.array([[0, 1, 0, 1],
                                            [1, 0, 1, 0],
                                               [0, 1, 0, 1],
                                               [1, 0, 1, 0]]))
        # Add edge weights to graph
        nx.set_edge_attributes(graph, {(0, 1): 4, (1, 2): 3, (2, 3): 2, (3, 0): 1}, "weight")
    elif topology_name == "7node":
        # 7 node ring
        graph = nx.from_numpy_array(jnp.array([[0, 1, 0, 0, 0, 0, 1],
                                               [1, 0, 1, 0, 0, 0, 0],
                                               [0, 1, 0, 1, 0, 0, 0],
                                               [0, 0, 1, 0, 1, 0, 0],
                                               [0, 0, 0, 1, 0, 1, 0],
                                               [0, 0, 0, 0, 1, 0, 1],
                                               [1, 0, 0, 0, 0, 1, 0]]))
        # Add edge weights to graph
        nx.set_edge_attributes(graph, {(0, 1): 4, (1, 2): 3, (2, 3): 2, (3, 4): 1, (4, 5): 2, (5, 6): 3, (6, 0): 4}, "weight")
    else:
        with open(topology_path / f"{topology_name}.json") as f:
            graph = nx.node_link_graph(json.load(f))
    return graph

mask_nodes(state, num_nodes)

Returns mask of valid actions for node selection. 1 for valid action, 0 for invalid action.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
num_nodes Scalar

Number of nodes

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
@partial(jax.jit, static_argnums=(1,))
def mask_nodes(state: EnvState, num_nodes: chex.Scalar) -> EnvState:
    """Returns mask of valid actions for node selection. 1 for valid action, 0 for invalid action.

    Args:
        state: Environment state
        num_nodes: Number of nodes

    Returns:
        state: Updated environment state
    """
    total_actions = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.action_counter, 1, 1))
    remaining_actions = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.action_counter, 2, 1))
    full_request = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.request_array, 0, 1))
    virtual_topology = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.request_array, 1, 1))
    request = jax.lax.dynamic_slice_in_dim(full_request, (remaining_actions - 1) * 2, 3)
    node_request_s = jax.lax.dynamic_slice_in_dim(request, 2, 1)
    node_request_d = jax.lax.dynamic_slice_in_dim(request, 0, 1)
    prev_action = jax.lax.dynamic_slice_in_dim(state.action_history, (remaining_actions) * 2, 3)
    prev_dest = jax.lax.dynamic_slice_in_dim(prev_action, 0, 1)
    node_indices = jnp.arange(0, num_nodes)
    # Get requested indices from request array virtual topology
    requested_indices = jax.lax.dynamic_slice_in_dim(virtual_topology, (remaining_actions-1)*2, 3)
    requested_index_d = jax.lax.dynamic_slice_in_dim(requested_indices, 0, 1)
    # Get index of previous selected node
    prev_selected_node = jnp.where(virtual_topology == requested_index_d, state.action_history, jnp.full(virtual_topology.shape, -1))
    # will be current index if node only occurs once in virtual topology or will be different index if occurs more than once
    prev_selected_index = jnp.argmax(prev_selected_node)
    prev_selected_node_d = jax.lax.dynamic_slice_in_dim(state.action_history, prev_selected_index, 1)

    # If first action, source and dest both to be assigned -> just mask all nodes based on resources
    # Thereafter, source must be previous dest. Dest can be any node (except previous allocations).
    state = state.replace(
        node_mask_s=jax.lax.cond(
            jnp.equal(remaining_actions, total_actions),
            lambda x: jnp.where(
                state.node_capacity_array >= node_request_s,
                x,
                jnp.zeros(num_nodes)
            ),
            lambda x: jnp.where(
                node_indices == prev_dest,
                x,
                jnp.zeros(num_nodes)
            ),
            jnp.ones(num_nodes),
        )
    )
    state = state.replace(
        node_mask_d=jnp.where(
            state.node_capacity_array >= node_request_d,
            jnp.ones(num_nodes),
            jnp.zeros(num_nodes)
        )
    )
    # If not first move, set node_mask_d to zero wherever node_mask_s is 1
    # to avoid same node selection for s and d
    state = state.replace(
        node_mask_d=jax.lax.cond(
            jnp.equal(remaining_actions, total_actions),
            lambda x: x,
            lambda x: jnp.where(
                state.node_mask_s == 1,
                jnp.zeros(num_nodes),
                x
            ),
            state.node_mask_d,
        )
    )

    def mask_previous_selections(i, val):
        # Disallow previously allocated nodes
        update_slice = lambda j, x: jax.lax.dynamic_update_slice_in_dim(x, jnp.array([0.]), j, axis=0)
        val = jax.lax.cond(
            i % 2 == 0,
            lambda x: update_slice(x[0][i], x[1]),  # i is node request index
            lambda x: update_slice(x[0][i+1], x[1]),  # i is slot request index (so add 1 to get next node)
            (state.action_history, val),
        )
        return val

    state = state.replace(
        node_mask_d=jax.lax.fori_loop(
            remaining_actions*2,
            state.action_history.shape[0]-1,
            mask_previous_selections,
            state.node_mask_d
        )
    )
    # If requested node index is new then disallow previously allocated nodes
    # If not new, then must match previously allocated node for that index
    state = state.replace(
        node_mask_d=jax.lax.cond(
            jnp.squeeze(prev_selected_node_d) >= 0,
            lambda x: jnp.where(
                node_indices == prev_selected_node_d,
                x[1],
                x[0],
            ),
            lambda x: x[2],
            (jnp.zeros(num_nodes), jnp.ones(num_nodes), state.node_mask_d),
        )
    )
    return state

mask_slots(state, params, request)

Returns binary mask of valid actions. 1 for valid action, 0 for invalid action.

  1. Check request for source and destination nodes
  2. For each path:
    • Get current slots on path (with padding on end to avoid out of bounds)
    • Get mask for required slots on path
    • Multiply through current slots with required slots mask to check if slots available on path
    • Remove padding from mask
    • Return path mask
  3. Update total mask with path mask
  4. If aggregate_slots > 1, aggregate slot mask to reduce action space

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required
request Array

Request array in format [source_node, data-rate, destination_node]

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
@partial(jax.jit, static_argnums=(1,))
def mask_slots(state: EnvState, params: EnvParams, request: chex.Array) -> EnvState:
    """Returns binary mask of valid actions. 1 for valid action, 0 for invalid action.

    1. Check request for source and destination nodes
    2. For each path:
        - Get current slots on path (with padding on end to avoid out of bounds)
        - Get mask for required slots on path
        - Multiply through current slots with required slots mask to check if slots available on path
        - Remove padding from mask
        - Return path mask
    3. Update total mask with path mask
    4. If aggregate_slots > 1, aggregate slot mask to reduce action space

    Args:
        state: Environment state
        params: Environment parameters
        request: Request array in format [source_node, data-rate, destination_node]

    Returns:
        state: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(request)
    init_mask = jnp.zeros((params.link_resources * params.k_paths))

    def mask_path(i, mask):
        # Get slots for path
        slots = get_path_slots(state.link_slot_array, params, nodes_sd, i)
        # Add padding to slots at end
        slots = jnp.concatenate((slots, jnp.ones(params.max_slots)))
        # Convert bandwidth to slots for each path
        spectral_efficiency = get_paths_se(params, nodes_sd)[i] if params.consider_modulation_format else 1
        requested_slots = required_slots(requested_datarate, spectral_efficiency, params.slot_size, guardband=params.guardband)
        # Get mask used to check if request will fit slots
        request_mask = get_request_mask(requested_slots[0], params)

        def check_slots_available(j, val):
            # Multiply through by request mask to check if slots available
            slot_sum = jnp.sum(request_mask * jax.lax.dynamic_slice(val, (j,), (params.max_slots,))) <= 0
            slot_sum = slot_sum.reshape((1,)).astype(jnp.float32)
            return jax.lax.dynamic_update_slice(val, slot_sum, (j,))

        # Mask out slots that are not valid
        path_mask = jax.lax.fori_loop(
            0,
            int(params.link_resources+1),  # No need to check last requested_slots-1 slots
            check_slots_available,
            slots,
        )
        # Cut off padding
        path_mask = jax.lax.dynamic_slice(path_mask, (0,), (params.link_resources,))
        # Update total mask with path mask
        mask = jax.lax.dynamic_update_slice(mask, path_mask, (i * params.link_resources,))
        return mask

    # Loop over each path
    link_slot_mask = jax.lax.fori_loop(0, params.k_paths, mask_path, init_mask)
    if params.aggregate_slots > 1:
        # Full link slot mask is used in process_path_action to get the correct slot from the aggregated slot action
        state = state.replace(full_link_slot_mask=link_slot_mask)
        link_slot_mask, _ = aggregate_slots(link_slot_mask.reshape(params.k_paths, -1), params)
        link_slot_mask = link_slot_mask.reshape(-1)
    state = state.replace(link_slot_mask=link_slot_mask)
    return state

mask_slots_rmsa_gn_model(state, params, request)

For use in RSAGNModelEnv. 1. For each path: 1.1 Get path slots 1.2 Get launch power

Parameters:

Name Type Description Default
state RSAGNModelEnvState

Environment state

 required
params RSAGNModelEnvParams

Environment parameters

 required
request Array

Request array in format [source_node, data-rate, destination_node]

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
@partial(jax.jit, static_argnums=(1,))
def mask_slots_rmsa_gn_model(state: RSAGNModelEnvState, params: RSAGNModelEnvParams, request: chex.Array) -> EnvState:
    """For use in RSAGNModelEnv.
    1. For each path:
        1.1 Get path slots
        1.2 Get launch power


    Args:
        state: Environment state
        params: Environment parameters
        request: Request array in format [source_node, data-rate, destination_node]

    Returns:
        state: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(request)
    init_mask = jnp.zeros((params.link_resources * params.k_paths))

    def mask_path(i, mask):
        path = get_paths(params, nodes_sd)[i]
        # Get slots for path
        slots = get_path_slots(state.link_slot_array, params, nodes_sd, i)
        # Add padding to slots at end
        # 0 means slot is free, 1 is occupied
        slots = jnp.concatenate((slots, jnp.ones(params.max_slots)))
        launch_power = get_launch_power(state, i, state.launch_power_array[i], params)
        lightpath_index = get_lightpath_index(params, nodes_sd, i)

        # This function checks through each available modulation format, checks the first and last available slots,
        # calculates the SNR, checks it meets the requirements, and returns the resulting mask
        def check_modulation_format(mod_format_index, init_path_mask):
            se = params.modulations_array.val[mod_format_index][1]
            req_slots = required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband)[0]
            bandwidth_per_subchannel = params.slot_size
            req_snr = params.modulations_array.val[mod_format_index][2] + params.snr_margin
            # Get mask used to check if request will fit slots
            request_mask = get_request_mask(req_slots, params)

            def check_slots_available(j, val):
                # Multiply through by request mask to check if slots available
                slot_sum = jnp.sum(request_mask * jax.lax.dynamic_slice(val, (j,), (params.max_slots,))) <= 0
                slot_sum = slot_sum.reshape((1,)).astype(jnp.float32)
                return jax.lax.dynamic_update_slice(val, slot_sum, (j,))

            # Mask out slots that are not valid
            slot_mask = jax.lax.fori_loop(
                0,
                int(params.link_resources + 1),  # No need to check last requested_slots-1 slots
                check_slots_available,
                slots,
            )
            # Cut off padding
            slot_mask = jax.lax.dynamic_slice(slot_mask, (0,), (params.link_resources,))
            # Check first and last available slots for suitability
            ff_path_mask = jnp.concatenate((slot_mask, jnp.ones((1,))), axis=0)
            lf_path_mask = jnp.concatenate((jnp.ones((1,)), slot_mask), axis=0)
            first_available_slot_index = jnp.argmax(ff_path_mask)
            last_available_slot_index = params.link_resources - jnp.argmax(jnp.flip(lf_path_mask)) - 1
            # Assign "req_slots" subchannels (each with bandwidth = slot width) for the first and last possible slots
            ff_temp_state = state.replace(
                channel_centre_bw_array=vmap_set_path_links(state.channel_centre_bw_array, path, first_available_slot_index, req_slots, bandwidth_per_subchannel),
                channel_power_array=vmap_set_path_links(state.channel_power_array, path, first_available_slot_index, req_slots, launch_power),
                path_index_array=vmap_set_path_links(state.path_index_array, path, first_available_slot_index, req_slots, lightpath_index),
                modulation_format_index_array=vmap_set_path_links(state.modulation_format_index_array, path, first_available_slot_index, req_slots, mod_format_index),
            )
            lf_temp_state = state.replace(
                channel_centre_bw_array=vmap_set_path_links(state.channel_centre_bw_array, path, last_available_slot_index, req_slots, bandwidth_per_subchannel),
                channel_power_array=vmap_set_path_links(state.channel_power_array, path, last_available_slot_index, req_slots, launch_power),
                path_index_array=vmap_set_path_links(state.path_index_array, path, last_available_slot_index, req_slots, lightpath_index),
                modulation_format_index_array=vmap_set_path_links(state.modulation_format_index_array, path, last_available_slot_index, req_slots, mod_format_index),
            )
            ff_temp_state = ff_temp_state.replace(link_snr_array=get_snr_link_array(ff_temp_state, params))
            lf_temp_state = lf_temp_state.replace(link_snr_array=get_snr_link_array(lf_temp_state, params))
            # Take the minimum value of SNR from all the subchannels
            ff_snr_value = get_minimum_snr_of_channels_on_path(
                ff_temp_state, path, first_available_slot_index, req_slots, params
            )
            lf_snr_value = get_minimum_snr_of_channels_on_path(
                lf_temp_state, path, last_available_slot_index, req_slots, params
            )
            # Check that other paths SNR is still sufficient (True if failure)
            ff_snr_check = 1 - check_action_rmsa_gn_model(ff_temp_state, None, params)
            lf_snr_check = 1 - check_action_rmsa_gn_model(lf_temp_state, None, params)
            ff_check = (ff_snr_value >= req_snr) * ff_snr_check
            lf_check = (lf_snr_value >= req_snr) * lf_snr_check

            slot_indices = jnp.arange(params.link_resources)
            mod_format_mask = jnp.where(slot_indices == first_available_slot_index, ff_check, False)
            mod_format_mask = jnp.where(slot_indices == last_available_slot_index, lf_check, mod_format_mask)
            path_mask = jnp.where(mod_format_mask, mod_format_index, init_path_mask).astype(jnp.float32)
            # jax.debug.print("ff_snr_check {}", ff_snr_check, ordered=True)
            # jax.debug.print("lf_snr_check {}", lf_snr_check, ordered=True)
            # jax.debug.print("ff_snr_value {}", ff_snr_value, ordered=True)
            # jax.debug.print("lf_snr_value {}", lf_snr_value, ordered=True)
            # jax.debug.print("first_available_slot_index {}", first_available_slot_index, ordered=True)
            # jax.debug.print("last_available_slot_index {}", last_available_slot_index, ordered=True)
            # jax.debug.print("req_snr {}", req_snr, ordered=True)
            # jax.debug.print("mod_format_mask {}", mod_format_mask, ordered=True)
            # jax.debug.print("path_mask {}", path_mask, ordered=True)
            return path_mask

        path_mask = jax.lax.fori_loop(0, params.modulations_array.val.shape[0], check_modulation_format, jnp.full((params.link_resources,), -1.))

        # Update total mask with path mask
        mask = jax.lax.dynamic_update_slice(mask, path_mask, (i * params.link_resources,))
        return mask

    # Loop over each path
    mod_format_mask = jax.lax.fori_loop(0, params.k_paths, mask_path, init_mask)
    link_slot_mask = jnp.where(mod_format_mask >= 0, 1.0, 0.0)
    if params.aggregate_slots > 1:
        # Full link slot mask is used in process_path_action to get the correct slot from the aggregated slot action
        state = state.replace(full_link_slot_mask=link_slot_mask)
        link_slot_mask, _ = aggregate_slots(link_slot_mask.reshape(params.k_paths, -1), params)
        link_slot_mask = link_slot_mask.reshape(-1)
    state = state.replace(
        link_slot_mask=link_slot_mask,
        mod_format_mask=mod_format_mask,
    )
    return state

mask_slots_rwalr(state, params, request)

For use in RWALightpathReuseEnv. Each lightpath has a maximum capacity defined in path_capacity_array. This is updated when a lightpath is assigned. If remaining path capacity is less than current request, corresponding link-slots are masked out. If link-slot is in use by another lightpath for a different source and destination node (even if not full) it is masked out. Step 1: - Mask out slots that are not valid based on path capacity (check link_capacity_array) Step 2: - Mask out slots that are not valid based on lightpath reuse (check path_index_array)

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required
request Array

Request array in format [source_node, data-rate, destination_node]

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
@partial(jax.jit, static_argnums=(1,))
def mask_slots_rwalr(state: EnvState, params: EnvParams, request: chex.Array) -> EnvState:
    """For use in RWALightpathReuseEnv.
    Each lightpath has a maximum capacity defined in path_capacity_array. This is updated when a lightpath is assigned.
    If remaining path capacity is less than current request, corresponding link-slots are masked out.
    If link-slot is in use by another lightpath for a different source and destination node (even if not full) it is masked out.
    Step 1:
    - Mask out slots that are not valid based on path capacity (check link_capacity_array)
    Step 2:
    - Mask out slots that are not valid based on lightpath reuse (check path_index_array)

    Args:
        state: Environment state
        params: Environment parameters
        request: Request array in format [source_node, data-rate, destination_node]

    Returns:
        state: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(request)
    init_mask = jnp.zeros((params.link_resources * params.k_paths))
    source, dest = nodes_sd
    path_start_index = get_path_indices(source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph).astype(jnp.int32)
    #jax.debug.print("path_start_index {}", path_start_index, ordered=True)
    #jax.debug.print("link_capacity_array {}", state.link_capacity_array, ordered=True)

    def mask_path(i, mask):
        # Step 1 - mask capacity
        capacity_mask = jnp.where(state.link_capacity_array < requested_datarate, 1., 0.)
        #jax.debug.print("capacity_mask {}", capacity_mask, ordered=True)
        capacity_slots = get_path_slots(capacity_mask, params, nodes_sd, i)
        #jax.debug.print("capacity_slots {}", capacity_slots, ordered=True)
        # Step 2 - mask lightpath reuse
        lightpath_index = path_start_index + i
        #jax.debug.print("lightpath_index {}", lightpath_index, ordered=True)
        lightpath_mask = jnp.where(state.path_index_array == lightpath_index, 0., 1.)  # Allow current lightpath
        #jax.debug.print("lightpath_mask {}", lightpath_mask, ordered=True)
        lightpath_mask = jnp.where(state.path_index_array == -1, 0., lightpath_mask)  # Allow empty slots
        #jax.debug.print("lightpath_mask {}", lightpath_mask, ordered=True)
        lightpath_slots = get_path_slots(lightpath_mask, params, nodes_sd, i)
        #jax.debug.print("lightpath_slots {}", lightpath_slots, ordered=True)
        # Step 3 combine masks
        path_mask = jnp.max(jnp.stack((capacity_slots, lightpath_slots)), axis=0)
        # Swap zeros for ones
        path_mask = jnp.where(path_mask == 0, 1., 0.)
        #jax.debug.print("path_mask {}", path_mask, ordered=True)
        mask = jax.lax.dynamic_update_slice(mask, path_mask, (i * params.link_resources,))
        return mask

    # Loop over each path
    link_slot_mask = jax.lax.fori_loop(0, params.k_paths, mask_path, init_mask)
    if params.aggregate_slots > 1:
        # Full link slot mask is used in process_path_action to get the correct slot from the aggregated slot action
        state = state.replace(full_link_slot_mask=link_slot_mask)
        link_slot_mask, _ = aggregate_slots(link_slot_mask.reshape(params.k_paths, -1), params)
        link_slot_mask = link_slot_mask.reshape(-1)
    state = state.replace(link_slot_mask=link_slot_mask)
    return state

normalise_traffic_matrix(traffic_matrix)

Normalise traffic matrix to sum to 1

Source code in xlron/environments/env_funcs.py 
580
581
582
583
def normalise_traffic_matrix(traffic_matrix):
    """Normalise traffic matrix to sum to 1"""
    traffic_matrix /= jnp.sum(traffic_matrix)
    return traffic_matrix

pad_array(array, fill_value)

Pad a ragged multidimensional array to rectangular shape. Used for training on multiple topologies. Source: https://codereview.stackexchange.com/questions/222623/pad-a-ragged-multidimensional-array-to-rectangular-shape

Parameters:

Name Type Description Default
array

array to pad

 required
fill_value

value to fill with

 required

Returns:

Name Type Description
result

padded array
Source code in xlron/environments/env_funcs.py 
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
def pad_array(array, fill_value):
    """
    Pad a ragged multidimensional array to rectangular shape.
    Used for training on multiple topologies.
    Source: https://codereview.stackexchange.com/questions/222623/pad-a-ragged-multidimensional-array-to-rectangular-shape

    Args:
        array: array to pad
        fill_value: value to fill with

    Returns:
        result: padded array
    """

    def get_dimensions(array, level=0):
        yield level, len(array)
        try:
            for row in array:
                yield from get_dimensions(row, level + 1)
        except TypeError: #not an iterable
            pass

    def get_max_shape(array):
        dimensions = defaultdict(int)
        for level, length in get_dimensions(array):
            dimensions[level] = max(dimensions[level], length)
        return [value for _, value in sorted(dimensions.items())]

    def iterate_nested_array(array, index=()):
        try:
            for idx, row in enumerate(array):
                yield from iterate_nested_array(row, (*index, idx))
        except TypeError: # final level
            yield (*index, slice(len(array))), array

    dimensions = get_max_shape(array)
    result = np.full(dimensions, fill_value)
    for index, value in iterate_nested_array(array):
        result[index] = value
    return result

path_action_only(topology_pattern, action_counter, remaining_actions)

This is to check if node has already been assigned, therefore just need to assign slots (n=0)

Parameters:

Name Type Description Default
topology_pattern Array

Topology pattern

 required
action_counter Array

Action counter

 required
remaining_actions Scalar

Remaining actions

 required

Returns:

Name Type Description
bool

True if only path action, False if node action
Source code in xlron/environments/env_funcs.py 
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
def path_action_only(topology_pattern: chex.Array, action_counter: chex.Array, remaining_actions: chex.Scalar):
    """This is to check if node has already been assigned, therefore just need to assign slots (n=0)

    Args:
        topology_pattern: Topology pattern
        action_counter: Action counter
        remaining_actions: Remaining actions

    Returns:
        bool: True if only path action, False if node action
    """
    # Get topology segment to be assigned e.g. [2,1,4]
    topology_segment = jax.lax.dynamic_slice(topology_pattern, ((remaining_actions-1)*2, ), (3, ))
    topology_indices = jnp.arange(topology_pattern.shape[0])
    # Check if the latest node in the segment is found in "prev_assigned_topology"
    new_node_to_be_assigned = topology_segment[0]
    prev_assigned_topology = jnp.where(topology_indices > (action_counter[-1]-1)*2, topology_pattern, 0)
    nodes_already_assigned_check = jnp.any(jnp.sum(jnp.where(prev_assigned_topology == new_node_to_be_assigned, 1, 0)) > 0)
    return nodes_already_assigned_check

poisson(key, lam, shape=(), dtype=dtypes.float_)

Sample Exponential random values with given shape and float dtype.

The values are distributed according to the probability density function:

.. math:: f(x) = \lambda e^{-\lambda x}

on the domain :math:0 \le x < \infty.

Args: key: a PRNG key used as the random key. lam: a positive float32 or float64 Tensor indicating the rate parameter shape: optional, a tuple of nonnegative integers representing the result shape. Default (). dtype: optional, a float dtype for the returned values (default float64 if jax_enable_x64 is true, otherwise float32).

Returns: A random array with the specified shape and dtype.

Source code in xlron/environments/env_funcs.py 
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
@partial(jax.jit, static_argnums=(1, 2, 3))
def poisson(key: Union[Array, prng.PRNGKeyArray],
            lam: ArrayLike,
            shape: Shape = (),
            dtype: DTypeLike = dtypes.float_) -> Array:
    r"""Sample Exponential random values with given shape and float dtype.

    The values are distributed according to the probability density function:

    .. math::
     f(x) = \lambda e^{-\lambda x}

    on the domain :math:`0 \le x < \infty`.

    Args:
    key: a PRNG key used as the random key.
    lam: a positive float32 or float64 `Tensor` indicating the rate parameter
    shape: optional, a tuple of nonnegative integers representing the result
      shape. Default ().
    dtype: optional, a float dtype for the returned values (default float64 if
      jax_enable_x64 is true, otherwise float32).

    Returns:
    A random array with the specified shape and dtype.
    """
    key, _ = jax._src.random._check_prng_key(key)
    if not dtypes.issubdtype(dtype, np.floating):
        raise ValueError(f"dtype argument to `exponential` must be a float "
                       f"dtype, got {dtype}")
    dtype = dtypes.canonicalize_dtype(dtype)
    shape = core.canonicalize_shape(shape)
    return _poisson(key, lam, shape, dtype)

process_path_action(state, params, path_action)

Process path action to get path index and initial slot index.

Parameters:

Name Type Description Default
state State

current state

 required
params Params

environment parameters

 required
path_action int

path action

 required

Returns:

Name Type Description
int (Array, Array)

path index
int (Array, Array)

initial slot index
Source code in xlron/environments/env_funcs.py 
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
@partial(jax.jit, static_argnums=(1,))
def process_path_action(state: EnvState, params: EnvParams, path_action: chex.Array) -> (chex.Array, chex.Array):
    """Process path action to get path index and initial slot index.

    Args:
        state (State): current state
        params (Params): environment parameters
        path_action (int): path action

    Returns:
        int: path index
        int: initial slot index
    """
    num_slot_actions = math.ceil(params.link_resources/params.aggregate_slots)
    path_index = jnp.floor(path_action / num_slot_actions).astype(jnp.int32).reshape(1)
    initial_aggregated_slot_index = jnp.mod(path_action, num_slot_actions).reshape(1)
    initial_slot_index = initial_aggregated_slot_index*params.aggregate_slots
    if params.aggregate_slots > 1:
        # Get the path mask do a dynamic slice and get the index of first unoccupied slot in the slice
        path_mask = jax.lax.dynamic_slice(state.full_link_slot_mask, path_index*params.link_resources, (params.link_resources,))
        path_mask_slice = jax.lax.dynamic_slice(path_mask, initial_slot_index, (params.aggregate_slots,))
        # Use argmax to get index of first 1 in slice of mask
        initial_slot_index = initial_slot_index + jnp.argmax(path_mask_slice)
    return path_index[0], initial_slot_index[0]

read_rsa_request(request_array)

Read RSA request from request array. Return source-destination nodes and bandwidth request.

Parameters:

Name Type Description Default
request_array Array

request array

 required

Returns:

Type Description
Tuple[Array, Array]

Tuple[chex.Array, chex.Array]: source-destination nodes and bandwidth request
Source code in xlron/environments/env_funcs.py 
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
def read_rsa_request(request_array: chex.Array) -> Tuple[chex.Array, chex.Array]:
    """Read RSA request from request array. Return source-destination nodes and bandwidth request.

    Args:
        request_array: request array

    Returns:
        Tuple[chex.Array, chex.Array]: source-destination nodes and bandwidth request
    """
    node_s = jax.lax.dynamic_slice(request_array, (0,), (1,))
    requested_datarate = jax.lax.dynamic_slice(request_array, (1,), (1,))
    node_d = jax.lax.dynamic_slice(request_array, (2,), (1,))
    nodes_sd = jnp.concatenate((node_s, node_d))
    return nodes_sd, requested_datarate

remove_expired_node_requests(state)

Check for values in node_departure_array that are less than the current time but greater than zero (negative time indicates the request is not yet finalised). If found, set to infinity in node_departure_array, set to zero in node_resource_array, and increase node_capacity_array by expired resources on each node.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
def remove_expired_node_requests(state: EnvState) -> EnvState:
    """Check for values in node_departure_array that are less than the current time but greater than zero
    (negative time indicates the request is not yet finalised).
    If found, set to infinity in node_departure_array, set to zero in node_resource_array, and increase
    node_capacity_array by expired resources on each node.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.node_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 < state.node_departure_array, mask, 0)
    expired_resources = jnp.sum(jnp.where(mask == 1, state.node_resource_array, 0), axis=1)
    state = state.replace(
        node_capacity_array=state.node_capacity_array + expired_resources,
        node_resource_array=jnp.where(mask == 1, 0, state.node_resource_array),
        node_departure_array=jnp.where(mask == 1, jnp.inf, state.node_departure_array)
    )
    return state

remove_expired_services_rmsa_gn_model(state)

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
def remove_expired_services_rmsa_gn_model(state: EnvState) -> EnvState:
    """

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.link_slot_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 <= state.link_slot_departure_array, mask, 0)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, 0, state.link_slot_array),
        link_slot_departure_array=jnp.where(mask == 1, 0, state.link_slot_departure_array),
        link_snr_array=jnp.where(mask == 1, 0., state.link_snr_array),
        path_index_array=jnp.where(mask == 1, -1, state.path_index_array),
        channel_centre_bw_array=jnp.where(mask == 1, 0, state.channel_centre_bw_array),
        channel_power_array=jnp.where(mask == 1, 0, state.channel_power_array),
        modulation_format_index_array=jnp.where(mask == 1, -1, state.modulation_format_index_array),
        path_index_array_prev=jnp.where(mask == 1, -1, state.path_index_array_prev),
        channel_centre_bw_array_prev=jnp.where(mask == 1, 0, state.channel_centre_bw_array_prev),
        channel_power_array_prev=jnp.where(mask == 1, 0, state.channel_power_array_prev),
        modulation_format_index_array_prev=jnp.where(mask == 1, -1, state.modulation_format_index_array_prev),
    )
    return state

remove_expired_services_rsa(state)

Check for values in link_slot_departure_array that are less than the current time but greater than zero (negative time indicates the request is not yet finalised). If found, set to zero in link_slot_array and link_slot_departure_array.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
def remove_expired_services_rsa(state: EnvState) -> EnvState:
    """Check for values in link_slot_departure_array that are less than the current time but greater than zero
    (negative time indicates the request is not yet finalised).
    If found, set to zero in link_slot_array and link_slot_departure_array.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.link_slot_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 <= state.link_slot_departure_array, mask, 0)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, 0, state.link_slot_array),
        link_slot_departure_array=jnp.where(mask == 1, 0, state.link_slot_departure_array)
    )
    return state

remove_expired_services_rsa_gn_model(state)

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
def remove_expired_services_rsa_gn_model(state: EnvState) -> EnvState:
    """

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.link_slot_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 <= state.link_slot_departure_array, mask, 0)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, 0, state.link_slot_array),
        link_slot_departure_array=jnp.where(mask == 1, 0, state.link_slot_departure_array),
        link_snr_array=jnp.where(mask == 1, 0., state.link_snr_array),
        path_index_array=jnp.where(mask == 1, -1, state.path_index_array),
        channel_centre_bw_array=jnp.where(mask == 1, 0, state.channel_centre_bw_array),
        channel_power_array=jnp.where(mask == 1, 0, state.channel_power_array),
        path_index_array_prev=jnp.where(mask == 1, -1, state.path_index_array_prev),
        channel_centre_bw_array_prev=jnp.where(mask == 1, 0, state.channel_centre_bw_array_prev),
        channel_power_array_prev=jnp.where(mask == 1, 0, state.channel_power_array_prev),
    )
    mask = jnp.where(state.active_lightpaths_array_departure < jnp.squeeze(state.current_time), 1, 0)
    # The active_lightpaths_array is set to -1 when the lightpath is not active
    # The active_lightpaths_array_departure is set to 0 when the lightpath is not active
    # (active_lightpaths_array is used to calculate the total throughput)
    mask = jnp.where(0 <= state.active_lightpaths_array_departure, mask, 0)
    state = state.replace(
        active_lightpaths_array=jnp.where(mask == 1, -1, state.active_lightpaths_array),
        active_lightpaths_array_departure=jnp.where(mask == 1, 0, state.active_lightpaths_array_departure),
    )
    return state

remove_expired_services_rwalr(state)

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
def remove_expired_services_rwalr(state: EnvState) -> EnvState:
    """

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.link_slot_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 <= state.link_slot_departure_array, mask, 0)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, 0, state.link_slot_array),
        link_slot_departure_array=jnp.where(mask == 1, 0, state.link_slot_departure_array),
        path_index_array=jnp.where(mask == 1, 0, state.path_index_array),
    )
    return state

required_slots(bitrate, se, channel_width, guardband=1)

Calculate required slots for a given bitrate and spectral efficiency.

Parameters:

Name Type Description Default
bit_rate float

Bit rate in Gbps

 required
se float

Spectral efficiency in bps/Hz

 required
channel_width float

Channel width in GHz

 required
guardband int

Guard band. Defaults to 1.

 1

Returns:

Name Type Description
int int

Required slots
Source code in xlron/environments/env_funcs.py 
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
@partial(jax.jit, static_argnums=(2, 3))
def required_slots(bitrate: float, se: int, channel_width: float, guardband: int = 1) -> int:
    """Calculate required slots for a given bitrate and spectral efficiency.

    Args:
        bit_rate (float): Bit rate in Gbps
        se (float): Spectral efficiency in bps/Hz
        channel_width (float): Channel width in GHz
        guardband (int, optional): Guard band. Defaults to 1.

    Returns:
        int: Required slots
    """
    # If bitrate is zero, then required slots should be zero
    return jnp.int32((jnp.ceil(bitrate/(se*channel_width))+guardband) * (1 - (bitrate == 0)))

set_c_l_band_gap(link_slot_array, params, val)

Set C+L band gap in link slot array Args: link_slot_array (chex.Array): Link slot array params (RSAGNModelEnvParams): Environment parameters val (int): Value to set Returns: chex.Array: Link slot array with C+L band gap

Source code in xlron/environments/env_funcs.py 
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
@partial(jax.jit, static_argnums=(1, 2))
def set_c_l_band_gap(link_slot_array: chex.Array, params: RSAGNModelEnvParams, val: int) -> chex.Array:
    """Set C+L band gap in link slot array
    Args:
        link_slot_array (chex.Array): Link slot array
        params (RSAGNModelEnvParams): Environment parameters
        val (int): Value to set
    Returns:
        chex.Array: Link slot array with C+L band gap
    """
    # Set C+L band gap
    gap = jnp.full((params.num_links, params.gap_width), val)
    link_slot_array = jax.lax.dynamic_update_slice(link_slot_array, gap, (0, params.gap_start))
    return link_slot_array

undo_action_rmsa_gn_model(state, params)

Undo action for RMSA GN model Args: state (EnvState): Environment state action (chex.Array): Action array params (EnvParams): Environment parameters Returns: EnvState: Updated environment state

Source code in xlron/environments/env_funcs.py 
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
@partial(jax.jit, static_argnums=(1,))
def undo_action_rmsa_gn_model(state: RSAGNModelEnvState, params: RSAGNModelEnvParams) -> EnvState:
    """Undo action for RMSA GN model
    Args:
        state (EnvState): Environment state
        action (chex.Array): Action array
        params (EnvParams): Environment parameters
    Returns:
        EnvState: Updated environment state
    """
    state = undo_action_rsa(state, params)  # Undo link_slot_array and link_slot_departure_array
    state = state.replace(
        link_slot_array=set_c_l_band_gap(state.link_slot_array, params, -1.),  # Set C+L band gap
        channel_centre_bw_array=state.channel_centre_bw_array_prev,
        path_index_array=state.path_index_array_prev,
        channel_power_array=state.channel_power_array_prev,
        modulation_format_index_array=state.modulation_format_index_array_prev,
    )
    return state

undo_action_rsa(state, params)

If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation. Partial resource allocation is indicated by negative time in link slot departure array. Check for values in link_slot_departure_array that are less than zero. If found, increase link_slot_array by +1 and link_slot_departure_array by current_time + holding_time of current request.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
@partial(jax.jit, donate_argnums=(0,))
def undo_action_rsa(state: EnvState, params: Optional[EnvParams]) -> EnvState:
    """If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation.
    Partial resource allocation is indicated by negative time in link slot departure array.
    Check for values in link_slot_departure_array that are less than zero.
    If found, increase link_slot_array by +1 and link_slot_departure_array by current_time + holding_time of current request.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    # If departure array is negative, then undo the action
    mask = jnp.where(state.link_slot_departure_array < 0, 1, 0)
    # If link slot array is < -1, then undo the action
    # (departure might be positive because existing service had holding time after current)
    # e.g. (time_in_array = t1 - t2) where t2 < t1 and t2 = current_time + holding_time
    # but link_slot_array = -2 due to double allocation, so undo the action
    mask = jnp.where(state.link_slot_array < -1, 1, mask)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, state.link_slot_array+1, state.link_slot_array),
        link_slot_departure_array=jnp.where(
            mask == 1,
            state.link_slot_departure_array + state.current_time + state.holding_time,
            state.link_slot_departure_array),
        total_bitrate=state.total_bitrate + read_rsa_request(state.request_array)[1][0]
    )
    return state

undo_action_rsa_gn_model(state, params)

Undo action for RSA GN model Args: state (EnvState): Environment state action (chex.Array): Action array params (EnvParams): Environment parameters Returns: EnvState: Updated environment state

Source code in xlron/environments/env_funcs.py 
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
@partial(jax.jit, static_argnums=(1,))
def undo_action_rsa_gn_model(state: RSAGNModelEnvState, params: RSAGNModelEnvParams) -> EnvState:
    """Undo action for RSA GN model
    Args:
        state (EnvState): Environment state
        action (chex.Array): Action array
        params (EnvParams): Environment parameters
    Returns:
        EnvState: Updated environment state
    """
    state = undo_action_rsa(state, params)  # Undo link_slot_array and link_slot_departure_array
    state = state.replace(
        link_slot_array=set_c_l_band_gap(state.link_slot_array, params, -1.),  # Set C+L band gap
        channel_centre_bw_array=state.channel_centre_bw_array_prev,
        path_index_array=state.path_index_array_prev,
        channel_power_array=state.channel_power_array_prev,
    )
    # If departure array is negative, then undo the action
    mask = jnp.where(state.active_lightpaths_array_departure < 0, 1, 0)
    state = state.replace(
        active_lightpaths_array=jnp.where(mask == 1, -1, state.active_lightpaths_array),
        active_lightpaths_array_departure=jnp.where(
            mask == 1,
            state.active_lightpaths_array_departure + state.current_time + state.holding_time,
            state.active_lightpaths_array_departure),
    )
    return state

undo_action_rwalr(state, params)

If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation. Partial resource allocation is indicated by negative time in link slot departure array. Check for values in link_slot_departure_array that are less than zero. If found, increase link_slot_array by +1 and link_slot_departure_array by current_time + holding_time of current request.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
@partial(jax.jit, donate_argnums=(0,))
def undo_action_rwalr(state: EnvState, params: Optional[EnvParams]) -> EnvState:
    """If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation.
    Partial resource allocation is indicated by negative time in link slot departure array.
    Check for values in link_slot_departure_array that are less than zero.
    If found, increase link_slot_array by +1 and link_slot_departure_array by current_time + holding_time of current request.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    # If departure array is negative, then undo the action
    mask = jnp.where(state.link_slot_departure_array < 0, 1, 0)
    # If link slot array is < -1, then undo the action
    # (departure might be positive because existing service had holding time after current)
    # e.g. (time_in_array = t1 - t2) where t2 < t1 and t2 = current_time + holding_time
    # but link_slot_array = -2 due to double allocation, so undo the action
    mask = jnp.where(state.link_slot_array < -1, 1, mask)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, state.link_slot_array + 1, state.link_slot_array),
        link_slot_departure_array=jnp.where(
            mask == 1,
            state.link_slot_departure_array + state.current_time + state.holding_time,
            state.link_slot_departure_array),
        total_bitrate=state.total_bitrate + read_rsa_request(state.request_array)[1][0]
    )
    return state

undo_node_action(state)

If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation. Partial resource allocation is indicated by negative time in node departure array. Check for values in node_departure_array that are less than zero. If found, set to infinity in node_departure_array, set to zero in node_resource_array, and increase node_capacity_array by expired resources on each node.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
@partial(jax.jit, donate_argnums=(0,))
def undo_node_action(state: EnvState) -> EnvState:
    """If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation.
    Partial resource allocation is indicated by negative time in node departure array.
    Check for values in node_departure_array that are less than zero.
    If found, set to infinity in node_departure_array, set to zero in node_resource_array, and increase
    node_capacity_array by expired resources on each node.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    # TODO - Check that node resource clash doesn't happen (so time is always negative after implementation)
    #  and undoing always succeeds with negative time
    mask = jnp.where(state.node_departure_array < 0, 1, 0)
    resources = jnp.sum(jnp.where(mask == 1, state.node_resource_array, 0), axis=1)
    state = state.replace(
        node_capacity_array=state.node_capacity_array + resources,
        node_resource_array=jnp.where(mask == 1, 0, state.node_resource_array),
        node_departure_array=jnp.where(mask == 1, jnp.inf, state.node_departure_array),
    )
    return state

update_action_history(action_history, action_counter, action)

Update action history by adding action to first available index starting from the end.

Parameters:

Name Type Description Default
action_history Array

Action history

 required
action_counter Array

Action counter

 required
action Array

Action to add to history

 required

Returns:

Type Description
Array

Updated action_history
Source code in xlron/environments/env_funcs.py 
850
851
852
853
854
855
856
857
858
859
860
861
def update_action_history(action_history: chex.Array, action_counter: chex.Array, action: chex.Array) -> chex.Array:
    """Update action history by adding action to first available index starting from the end.

    Args:
        action_history: Action history
        action_counter: Action counter
        action: Action to add to history

    Returns:
        Updated action_history
    """
    return jax.lax.dynamic_update_slice(action_history, jnp.flip(action, axis=0).astype(jnp.int32), ((action_counter[-1]-1)*2,))

update_active_lightpaths_array(state, path_index)

Update active lightpaths array with new path index. Find the first index of the array with value -1 and replace with path index. Args: state (RSAGNModelEnvState): Environment state path_index (int): Path index to add to active lightpaths array Returns: jnp.array: Updated active lightpaths array

Source code in xlron/environments/env_funcs.py 
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
def update_active_lightpaths_array(state: RSAGNModelEnvState, path_index: int) -> chex.Array:
    """Update active lightpaths array with new path index.
    Find the first index of the array with value -1 and replace with path index.
    Args:
        state (RSAGNModelEnvState): Environment state
        path_index (int): Path index to add to active lightpaths array
    Returns:
        jnp.array: Updated active lightpaths array
    """
    first_empty_index = jnp.argmin(state.active_lightpaths_array)
    return jax.lax.dynamic_update_slice(state.active_lightpaths_array, jnp.array([path_index]), (first_empty_index,))

update_active_lightpaths_array_departure(state, time)

Update active lightpaths array with new path index. Find the first index of the array with value -1 and replace with path index. Args: state (RSAGNModelEnvState): Environment state time (float): Departure time Returns: jnp.array: Updated active lightpaths array

Source code in xlron/environments/env_funcs.py 
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
def update_active_lightpaths_array_departure(state: RSAGNModelEnvState, time: float) -> chex.Array:
    """Update active lightpaths array with new path index.
    Find the first index of the array with value -1 and replace with path index.
    Args:
        state (RSAGNModelEnvState): Environment state
        time (float): Departure time
    Returns:
        jnp.array: Updated active lightpaths array
    """
    first_empty_index = jnp.argmin(state.active_lightpaths_array)
    return jax.lax.dynamic_update_slice(state.active_lightpaths_array_departure, time, (first_empty_index,))

update_graph_tuple(state, params)

Update graph tuple for use with Jraph GNNs. Edge and node features are updated from link_slot_array and node_capacity_array respectively. Global features are updated as request_array. Args: state (EnvState): Environment state params (EnvParams): Environment parameters Returns: state (EnvState): Environment state with updated graph tuple

Source code in xlron/environments/env_funcs.py 
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
def update_graph_tuple(state: EnvState, params: EnvParams):
    """Update graph tuple for use with Jraph GNNs.
    Edge and node features are updated from link_slot_array and node_capacity_array respectively.
    Global features are updated as request_array.
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
    Returns:
        state (EnvState): Environment state with updated graph tuple
    """
    # Get source and dest from request array
    source_dest, _ = read_rsa_request(state.request_array)
    source, dest = source_dest[0], source_dest[2]
    # Current request as global feature
    globals = jnp.reshape(state.request_array, (1, -1))
    # One-hot encode source and destination
    source_dest_features = jnp.zeros((params.num_nodes, 2))
    source_dest_features = source_dest_features.at[source.astype(jnp.int32), 0].set(1)
    source_dest_features = source_dest_features.at[dest.astype(jnp.int32), 1].set(-1)
    spectral_features = state.graph.nodes[..., :3]

    if params.__class__.__name__ in ["RSAGNModelEnvParams", "RMSAGNModelEnvParams"]:
        # Normalize by max parameters (converted to linear units)
        max_power = isrs_gn_model.from_dbm(params.max_power)
        normalized_power = jnp.round(state.channel_power_array / max_power, 3)
        max_snr = isrs_gn_model.from_db(params.max_snr)
        normalized_snr = jnp.round(state.link_snr_array / max_snr, 3)
        edge_features = jnp.stack([normalized_snr, normalized_power], axis=-1)
        node_features = jnp.concatenate([spectral_features, source_dest_features], axis=-1)
    elif params.__class__.__name__ == "VONEEnvParams":
        edge_features = state.link_slot_array
        node_features = getattr(state, "node_capacity_array", jnp.zeros(params.num_nodes))
        node_features = node_features.reshape(-1, 1)
        node_features = jnp.concatenate([node_features, spectral_features, source_dest_features], axis=-1)
    else:
        edge_features = state.link_slot_array
        node_features = jnp.concatenate([spectral_features, source_dest_features], axis=-1)

    if params.disable_node_features:
        node_features = jnp.zeros((1,))

    edge_features = edge_features if params.directed_graph else jnp.repeat(edge_features, 2, axis=0)
    graph = state.graph._replace(nodes=node_features, edges=edge_features, globals=globals)
    state = state.replace(graph=graph)
    return state

update_node_array(node_indices, array, node, request)

Used to udated selected_nodes array with new requested resources on each node, for use in

Source code in xlron/environments/env_funcs.py 
973
974
975
def update_node_array(node_indices, array, node, request):
    """Used to udated selected_nodes array with new requested resources on each node, for use in """
    return jnp.where(node_indices == node, array-request, array)

Set relevant slots along links in path to val.

Parameters:

Name Type Description Default
link_slot_array Array

Link slot array

 required
path Array

Path (row from path-link array that indicates links used by path)

 required
initial_slot int

Initial slot

 required
num_slots int

Number of slots

 required
value int

Value to set on link slot array

 required

Returns:

Type Description
Array

Updated link slot array
Source code in xlron/environments/env_funcs.py 
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
@jax.jit
def vmap_set_path_links(link_slot_array: chex.Array, path: chex.Array, initial_slot: int, num_slots: int, value: int) -> chex.Array:
    """Set relevant slots along links in path to val.

    Args:
        link_slot_array: Link slot array
        path: Path (row from path-link array that indicates links used by path)
        initial_slot: Initial slot
        num_slots: Number of slots
        value: Value to set on link slot array

    Returns:
        Updated link slot array
    """
    return jax.vmap(set_path, in_axes=(0, 0, None, None, None))(link_slot_array, path, initial_slot, num_slots, value)

vmap_update_node_departure(node_departure_array, selected_nodes, value)

Called when implementing node action. Sets request departure time ("value") in place of first "inf" i.e. unoccupied index on node departure array for selected nodes.

Parameters:

Name Type Description Default
node_departure_array Array

(N x R) Node departure array

 required
selected_nodes Array

(N x 1) Selected nodes (non-zero value on selected node indices)

 required
value int

Value to set on node departure array

 required

Returns:

Type Description
Array

Updated node departure array
Source code in xlron/environments/env_funcs.py 
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
@jax.jit
def vmap_update_node_departure(node_departure_array: chex.Array, selected_nodes: chex.Array, value: int) -> chex.Array:
    """Called when implementing node action.
    Sets request departure time ("value") in place of first "inf" i.e. unoccupied index on node departure array for selected nodes.

    Args:
        node_departure_array: (N x R) Node departure array
        selected_nodes: (N x 1) Selected nodes (non-zero value on selected node indices)
        value: Value to set on node departure array

    Returns:
        Updated node departure array
    """
    first_inf_indices = jnp.argmax(node_departure_array, axis=1)
    return jax.vmap(update_selected_node_departure, in_axes=(0, 0, 0, None))(node_departure_array, selected_nodes, first_inf_indices, value)

vmap_update_node_resources(node_resource_array, selected_nodes)

Called when implementing node action. Sets requested node resources on selected nodes in place of first "zero" i.e. unoccupied index on node resource array for selected nodes.

Parameters:

Name Type Description Default
node_resource_array

(N x R) Node resource array

 required
selected_nodes

(N x 1) Requested resources on selected nodes

 required

Returns:

Type Description

Updated node resource array
Source code in xlron/environments/env_funcs.py 
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
@jax.jit
def vmap_update_node_resources(node_resource_array, selected_nodes):
    """Called when implementing node action.
    Sets requested node resources on selected nodes in place of first "zero" i.e.
    unoccupied index on node resource array for selected nodes.

    Args:
        node_resource_array: (N x R) Node resource array
        selected_nodes: (N x 1) Requested resources on selected nodes

    Returns:
        Updated node resource array
    """
    first_zero_indices = jnp.argmin(node_resource_array, axis=1)
    return jax.vmap(update_selected_node_resources, in_axes=(0, 0, 0))(node_resource_array, selected_nodes, first_zero_indices)

Update relevant slots along links in path to current_val - val.

Parameters:

Name Type Description Default
link_slot_array Array

Link slot array

 required
path Array

Path (row from path-link array that indicates links used by path)

 required
initial_slot int

Initial slot

 required
num_slots int

Number of slots

 required
value int

Value to subtract from link slot array

 required

Returns:

Type Description
Array

Updated link slot array
Source code in xlron/environments/env_funcs.py 
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
@jax.jit
def vmap_update_path_links(link_slot_array: chex.Array, path: chex.Array, initial_slot: int, num_slots: int, value: int) -> chex.Array:
    """Update relevant slots along links in path to current_val - val.

    Args:
        link_slot_array: Link slot array
        path: Path (row from path-link array that indicates links used by path)
        initial_slot: Initial slot
        num_slots: Number of slots
        value: Value to subtract from link slot array

    Returns:
        Updated link slot array
    """
    return jax.vmap(update_path, in_axes=(0, 0, None, None, None))(link_slot_array, path, initial_slot, num_slots, value)

Models

ActorCriticGNN

Bases: Module

Combine the GNN actor and critic networks into a single class

Source code in xlron/models/models.py 
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
class ActorCriticGNN(nn.Module):
    """Combine the GNN actor and critic networks into a single class"""
    activation: str = "tanh"
    num_layers: int = 2
    num_units: int = 64
    gnn_latent: int = 128
    message_passing_steps: int = 1
    output_edges_size_critic: int = 10
    output_nodes_size_critic: int = 0
    output_globals_size_critic: int = 1
    output_edges_size_actor: int = 10
    output_nodes_size_actor: int = 0
    output_globals_size_actor: int = 1
    gnn_mlp_layers: int = 1
    use_attention: bool = True
    normalise_by_link_length: bool = True  # Normalise the processed edge features by the link length
    gnn_layer_norm: bool = True
    mlp_layer_norm: bool = False
    vmap: bool = True
    temperature: float = 1.0
    # Launch power specific parameters
    min_power_dbm: float = 0.0
    max_power_dbm: float = 2.0
    step_power_dbm: float = 0.1
    discrete: bool = True
    # Beta distribution parameters (used only if discrete=False)
    min_concentration: float = 0.1
    max_concentration: float = 20.0
    epsilon: float = 1e-6
    # Bools to determine which actions to output
    output_path: bool = True
    output_power: bool = True

    @property
    def num_power_levels(self):
        """Calculate number of power levels dynamically"""
        return int((self.max_power_dbm - self.min_power_dbm) / self.step_power_dbm) + 1

    @property
    def power_levels(self):
        """Calculate power levels dynamically"""
        return jnp.linspace(self.min_power_dbm, self.max_power_dbm, self.num_power_levels)

    @nn.compact
    def __call__(self, state: EnvState, params: EnvParams):
        actor = ActorGNN(
            num_layers=self.num_layers,
            num_units=self.num_units,
            gnn_latent=self.gnn_latent,
            message_passing_steps=self.message_passing_steps,
            output_edges_size=self.output_edges_size_actor,
            output_nodes_size=self.output_nodes_size_actor,
            output_globals_size=self.output_globals_size_actor,
            gnn_mlp_layers=self.gnn_mlp_layers,
            use_attention=self.use_attention,
            normalise_by_link_length=self.normalise_by_link_length,
            gnn_layer_norm=self.gnn_layer_norm,
            mlp_layer_norm=self.mlp_layer_norm,
            temperature=self.temperature,
            min_power_dbm=self.min_power_dbm,
            max_power_dbm=self.max_power_dbm,
            step_power_dbm=self.step_power_dbm,
            discrete=self.discrete,
            min_concentration=self.min_concentration,
            max_concentration=self.max_concentration,
            epsilon=self.epsilon,
        )
        critic = CriticGNN(
            activation=self.activation,
            num_layers=self.num_layers,
            num_units=self.num_units,
            gnn_latent=self.gnn_latent,
            message_passing_steps=self.message_passing_steps,
            output_edges_size=self.output_edges_size_critic,
            output_nodes_size=self.output_nodes_size_critic,
            output_globals_size=self.output_globals_size_critic,
            gnn_mlp_layers=self.gnn_mlp_layers,
            use_attention=self.use_attention,
            normalise_by_link_length=self.normalise_by_link_length,
            gnn_layer_norm=self.gnn_layer_norm,
            mlp_layer_norm=self.mlp_layer_norm,
        )

        if self.vmap:
            actor = jax.vmap(actor, in_axes=(0, None))
            critic = jax.vmap(critic, in_axes=(0, None))
        actor_out = actor(state, params)
        critic_out = critic(state, params)
        return actor_out, critic_out

    def sample_action_path(self, seed, dist, log_prob=False, deterministic=False):
        """Sample an action from the distribution."""
        action = jnp.argmax(dist.probs()) if deterministic else dist.sample(seed=seed)
        if log_prob:
            return action, dist.log_prob(action)
        return action

    def sample_action_power(self, seed, dist, log_prob=False, deterministic=False):
        """Sample an action and convert to power level"""
        if self.discrete:
            if deterministic:
                # Take most probable action
                raw_action = dist.mode()
            else:
                # Sample from distribution
                raw_action = dist.sample(seed=seed)
            processed_action = self.power_levels[raw_action]
        else:
            if deterministic:
                # Use mean of Beta distribution for deterministic action
                mean = dist.alpha / (dist.alpha + dist.beta)
                raw_action = jnp.clip(mean, self.epsilon, 1.0 - self.epsilon)
            else:
                # Sample from Beta (clipping to avoid edge values with undefined gradient) and scale to power range
                raw_action = jnp.clip(dist.sample(seed=seed), self.epsilon, 1.0 - self.epsilon)
            processed_action = self.min_power_dbm + raw_action * (self.max_power_dbm - self.min_power_dbm)
        processed_action = from_dbm(processed_action)
        if log_prob:
            return processed_action, dist.log_prob(raw_action)
        return processed_action

    def sample_action_path_power(self, seed, dist, log_prob=False, deterministic=False):
        """Sample an action from the distributions.
        This assumes dist is a tuple of path and power distributions."""
        path_action = self.sample_action_path(seed, dist[0], log_prob=log_prob, deterministic=deterministic)
        power_action  = self.sample_action_power(seed, dist[1], log_prob=log_prob, deterministic=deterministic)
        if log_prob:
            return path_action[0], power_action[0], path_action[1]+power_action[1]
        return path_action, power_action

    def sample_action(self, seed, dist, log_prob=False, deterministic=False):
        """Sample an action from the distributions.
        This assumes dist is a tuple of path and power distributions OR just the appropriate distribution."""
        if self.output_path and self.output_power:
            return self.sample_action_path_power(seed, dist, log_prob=log_prob, deterministic=deterministic)
        elif self.output_path:
            return self.sample_action_path(seed, dist, log_prob=log_prob, deterministic=deterministic)
        elif self.output_power:
            return self.sample_action_power(seed, dist, log_prob=log_prob, deterministic=deterministic)
        else:
            raise ValueError("No action type specified for sampling.")

num_power_levels property

Calculate number of power levels dynamically

power_levels property

Calculate power levels dynamically

sample_action(seed, dist, log_prob=False, deterministic=False)

Sample an action from the distributions. This assumes dist is a tuple of path and power distributions OR just the appropriate distribution.

Source code in xlron/models/models.py 
708
709
710
711
712
713
714
715
716
717
718
def sample_action(self, seed, dist, log_prob=False, deterministic=False):
    """Sample an action from the distributions.
    This assumes dist is a tuple of path and power distributions OR just the appropriate distribution."""
    if self.output_path and self.output_power:
        return self.sample_action_path_power(seed, dist, log_prob=log_prob, deterministic=deterministic)
    elif self.output_path:
        return self.sample_action_path(seed, dist, log_prob=log_prob, deterministic=deterministic)
    elif self.output_power:
        return self.sample_action_power(seed, dist, log_prob=log_prob, deterministic=deterministic)
    else:
        raise ValueError("No action type specified for sampling.")

sample_action_path(seed, dist, log_prob=False, deterministic=False)

Sample an action from the distribution.

Source code in xlron/models/models.py 
668
669
670
671
672
673
def sample_action_path(self, seed, dist, log_prob=False, deterministic=False):
    """Sample an action from the distribution."""
    action = jnp.argmax(dist.probs()) if deterministic else dist.sample(seed=seed)
    if log_prob:
        return action, dist.log_prob(action)
    return action

sample_action_path_power(seed, dist, log_prob=False, deterministic=False)

Sample an action from the distributions. This assumes dist is a tuple of path and power distributions.

Source code in xlron/models/models.py 
699
700
701
702
703
704
705
706
def sample_action_path_power(self, seed, dist, log_prob=False, deterministic=False):
    """Sample an action from the distributions.
    This assumes dist is a tuple of path and power distributions."""
    path_action = self.sample_action_path(seed, dist[0], log_prob=log_prob, deterministic=deterministic)
    power_action  = self.sample_action_power(seed, dist[1], log_prob=log_prob, deterministic=deterministic)
    if log_prob:
        return path_action[0], power_action[0], path_action[1]+power_action[1]
    return path_action, power_action

sample_action_power(seed, dist, log_prob=False, deterministic=False)

Sample an action and convert to power level

Source code in xlron/models/models.py 
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
def sample_action_power(self, seed, dist, log_prob=False, deterministic=False):
    """Sample an action and convert to power level"""
    if self.discrete:
        if deterministic:
            # Take most probable action
            raw_action = dist.mode()
        else:
            # Sample from distribution
            raw_action = dist.sample(seed=seed)
        processed_action = self.power_levels[raw_action]
    else:
        if deterministic:
            # Use mean of Beta distribution for deterministic action
            mean = dist.alpha / (dist.alpha + dist.beta)
            raw_action = jnp.clip(mean, self.epsilon, 1.0 - self.epsilon)
        else:
            # Sample from Beta (clipping to avoid edge values with undefined gradient) and scale to power range
            raw_action = jnp.clip(dist.sample(seed=seed), self.epsilon, 1.0 - self.epsilon)
        processed_action = self.min_power_dbm + raw_action * (self.max_power_dbm - self.min_power_dbm)
    processed_action = from_dbm(processed_action)
    if log_prob:
        return processed_action, dist.log_prob(raw_action)
    return processed_action

ActorCriticMLP

Bases: Module

Source code in xlron/models/models.py 
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
class ActorCriticMLP(nn.Module):
    action_dim: Sequence[int]
    activation: str = "tanh"
    num_layers: int = 2
    num_units: int = 64
    layer_norm: bool = False
    temperature: float = 1.0

    @nn.compact
    def __call__(self, x):
        actor_mean = make_dense_layers(x, self.num_units, self.num_layers, self.activation)
        stacked_logits = []
        actor_mean_dim = nn.Dense(
            self.action_dim, kernel_init=orthogonal(0.01), bias_init=constant(0.0)
        )(actor_mean)
        logits = actor_mean_dim / self.temperature
        stacked_logits.append(logits)

        # If there are multiple action dimensions, concatenate the logits
        action_dist = distrax.Categorical(logits=logits)

        critic = make_dense_layers(x, self.num_units, self.num_layers, self.activation, layer_norm=self.layer_norm)
        critic = nn.Dense(1, kernel_init=orthogonal(1.0), bias_init=constant(0.0))(
            critic
        )

        return action_dist, jnp.squeeze(critic, axis=-1)

    def sample_action(self, seed,  dist, log_prob=False, deterministic=False):
        """Sample an action from the distribution"""
        action = jnp.argmax(dist.probs()) if deterministic else dist.sample(seed=seed)
        if log_prob:
            return action, dist.log_prob(action)
        return action

sample_action(seed, dist, log_prob=False, deterministic=False)

Sample an action from the distribution

Source code in xlron/models/models.py 
117
118
119
120
121
122
def sample_action(self, seed,  dist, log_prob=False, deterministic=False):
    """Sample an action from the distribution"""
    action = jnp.argmax(dist.probs()) if deterministic else dist.sample(seed=seed)
    if log_prob:
        return action, dist.log_prob(action)
    return action

ActorGNN

Bases: Module

Actor network for PPO algorithm. Takes the current state and returns a distrax.Categorical distribution over actions.

Source code in xlron/models/models.py 
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
class ActorGNN(nn.Module):
    """
    Actor network for PPO algorithm. Takes the current state and returns a distrax.Categorical distribution
    over actions.
    """
    activation: str = "tanh"
    num_layers: int = 2
    num_units: int = 64
    gnn_latent: int = 128
    dropout_rate: float = 0
    deterministic: bool = False
    message_passing_steps: int = 1
    output_edges_size: int = 10
    output_nodes_size: int = 0
    output_globals_size: int = 0
    gnn_mlp_layers: int = 1
    use_attention: bool = True
    normalise_by_link_length: bool = True  # Normalise the processed edge features by the link length
    gnn_layer_norm: bool = True
    mlp_layer_norm: bool = False
    temperature: float = 1.0
    # Launch power specific parameters
    min_power_dbm: float = 0.0
    max_power_dbm: float = 2.0
    step_power_dbm: float = 0.1
    discrete: bool = True
    # Beta distribution parameters (used only if discrete=False)
    min_concentration: float = 0.1
    max_concentration: float = 20.0
    epsilon: float = 1e-6

    @property
    def num_power_levels(self):
        """Calculate number of power levels dynamically"""
        return int((self.max_power_dbm - self.min_power_dbm) / self.step_power_dbm) + 1

    @property
    def power_levels(self):
        """Calculate power levels dynamically"""
        return jnp.linspace(self.min_power_dbm, self.max_power_dbm, self.num_power_levels)

    @nn.compact
    def __call__(self, state: EnvState, params: EnvParams):
        """
        The ActorGNN network takes the current network state in the form of a GraphTuple and returns
        a distrax.Categorical distribution over actions.
        The graph is processed by a GraphNet module, and the resulting graph is indexed to construct a matrix of
        the edge features. The edge features are then normalised by the link length from the environment parameters,
        and the current request is read from the request array.
        The request is used to retrieve the edge features from the edge_features array for the corresponding
        shortest k-paths. The edge features are aggregated for each path according to the "agg_func" e.g. sum,
        and the action distribution array is updated.
        Returns a distrax.Categorical distribution, from which actions can be sampled.

        :param state: EnvState
        :param params: EnvParams

        :return: distrax.Categorical distribution over actions
        """
        processed_graph = GraphNet(
            latent_size=self.gnn_latent,
            message_passing_steps=self.message_passing_steps,
            output_edges_size=self.output_edges_size,
            output_nodes_size=self.output_nodes_size,
            output_globals_size=self.output_globals_size,
            num_mlp_layers=self.gnn_mlp_layers,
            use_attention=self.use_attention,
            gnn_layer_norm=self.gnn_layer_norm,
            mlp_layer_norm=self.mlp_layer_norm,
        )(state.graph)

        # Index edge features to resemble the link-slot array
        edge_features = processed_graph.edges if params.directed_graph else processed_graph.edges[:len(processed_graph.edges) // 2]
        # Normalise features by normalised link length from state.link_length_array
        if self.normalise_by_link_length:
            edge_features = edge_features * (params.link_length_array.val/jnp.sum(params.link_length_array.val))
        # Get the current request and initialise array of action distributions per path
        nodes_sd, requested_bw = read_rsa_request(state.request_array)
        init_action_array = jnp.zeros(params.k_paths * self.output_edges_size)

        # Define a body func to retrieve path slots and update action array
        def get_path_action_dist(i, action_array):
            # Get the processed graph edge features corresponding to the i-th path
            path_features = get_path_slots(edge_features, params, nodes_sd, i, agg_func="sum")
            # Update the action array with the path features
            action_array = jax.lax.dynamic_update_slice(action_array, path_features, (i * self.output_edges_size,))
            return action_array

        path_action_logits = jax.lax.fori_loop(0, params.k_paths, get_path_action_dist, init_action_array)
        path_action_logits = jnp.reshape(path_action_logits, (-1,)) / self.temperature

        power_action_dist = None
        mlp_features = [self.num_units] * self.num_layers
        output_size = self.num_power_levels if self.discrete else 2
        path_mlp = MLP(
            mlp_features + [output_size],
            dropout_rate=self.dropout_rate,
            deterministic=self.deterministic,
            layer_norm=self.mlp_layer_norm,
        )
        if params.__class__.__name__ == "RSAGNModelEnvParams":
            if self.output_globals_size > 0:
                power_logits = processed_graph.globals.reshape((-1,)) / self.temperature
            else:
                init_feature_array = jnp.zeros((params.k_paths, edge_features.shape[1]))
                # Define a body func to retrieve path slots and update action array
                def get_power_action_dist(i, feature_array):
                    # Get the processed graph edge features corresponding to the i-th path
                    path_features = get_path_slots(edge_features, params, nodes_sd, i, agg_func="sum").reshape((1, -1))
                    # Update the array with the path features
                    action_array = jax.lax.dynamic_update_slice(feature_array, path_features,(i, 0))
                    return action_array
                path_feature_batch = jax.lax.fori_loop(0, params.k_paths, get_power_action_dist, init_feature_array)
                power_logits = path_mlp(path_feature_batch)
            if self.discrete:
                power_action_dist = distrax.Categorical(logits=power_logits)
            else:
                alpha = self.min_concentration + jax.nn.softplus(power_logits) * (
                        self.max_concentration - self.min_concentration
                )
                beta = self.min_concentration + jax.nn.softplus(power_logits) * (
                        self.max_concentration - self.min_concentration
                )
                power_action_dist = distrax.Beta(alpha, beta)

        # Return a distrax.Categorical distribution over actions (which can be masked later)
        path_action_dist = distrax.Categorical(logits=path_action_logits)
        return (path_action_dist, power_action_dist)

num_power_levels property

Calculate number of power levels dynamically

power_levels property

Calculate power levels dynamically

__call__(state, params)

The ActorGNN network takes the current network state in the form of a GraphTuple and returns a distrax.Categorical distribution over actions. The graph is processed by a GraphNet module, and the resulting graph is indexed to construct a matrix of the edge features. The edge features are then normalised by the link length from the environment parameters, and the current request is read from the request array. The request is used to retrieve the edge features from the edge_features array for the corresponding shortest k-paths. The edge features are aggregated for each path according to the "agg_func" e.g. sum, and the action distribution array is updated. Returns a distrax.Categorical distribution, from which actions can be sampled.

:param state: EnvState :param params: EnvParams

:return: distrax.Categorical distribution over actions

Source code in xlron/models/models.py 
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
@nn.compact
def __call__(self, state: EnvState, params: EnvParams):
    """
    The ActorGNN network takes the current network state in the form of a GraphTuple and returns
    a distrax.Categorical distribution over actions.
    The graph is processed by a GraphNet module, and the resulting graph is indexed to construct a matrix of
    the edge features. The edge features are then normalised by the link length from the environment parameters,
    and the current request is read from the request array.
    The request is used to retrieve the edge features from the edge_features array for the corresponding
    shortest k-paths. The edge features are aggregated for each path according to the "agg_func" e.g. sum,
    and the action distribution array is updated.
    Returns a distrax.Categorical distribution, from which actions can be sampled.

    :param state: EnvState
    :param params: EnvParams

    :return: distrax.Categorical distribution over actions
    """
    processed_graph = GraphNet(
        latent_size=self.gnn_latent,
        message_passing_steps=self.message_passing_steps,
        output_edges_size=self.output_edges_size,
        output_nodes_size=self.output_nodes_size,
        output_globals_size=self.output_globals_size,
        num_mlp_layers=self.gnn_mlp_layers,
        use_attention=self.use_attention,
        gnn_layer_norm=self.gnn_layer_norm,
        mlp_layer_norm=self.mlp_layer_norm,
    )(state.graph)

    # Index edge features to resemble the link-slot array
    edge_features = processed_graph.edges if params.directed_graph else processed_graph.edges[:len(processed_graph.edges) // 2]
    # Normalise features by normalised link length from state.link_length_array
    if self.normalise_by_link_length:
        edge_features = edge_features * (params.link_length_array.val/jnp.sum(params.link_length_array.val))
    # Get the current request and initialise array of action distributions per path
    nodes_sd, requested_bw = read_rsa_request(state.request_array)
    init_action_array = jnp.zeros(params.k_paths * self.output_edges_size)

    # Define a body func to retrieve path slots and update action array
    def get_path_action_dist(i, action_array):
        # Get the processed graph edge features corresponding to the i-th path
        path_features = get_path_slots(edge_features, params, nodes_sd, i, agg_func="sum")
        # Update the action array with the path features
        action_array = jax.lax.dynamic_update_slice(action_array, path_features, (i * self.output_edges_size,))
        return action_array

    path_action_logits = jax.lax.fori_loop(0, params.k_paths, get_path_action_dist, init_action_array)
    path_action_logits = jnp.reshape(path_action_logits, (-1,)) / self.temperature

    power_action_dist = None
    mlp_features = [self.num_units] * self.num_layers
    output_size = self.num_power_levels if self.discrete else 2
    path_mlp = MLP(
        mlp_features + [output_size],
        dropout_rate=self.dropout_rate,
        deterministic=self.deterministic,
        layer_norm=self.mlp_layer_norm,
    )
    if params.__class__.__name__ == "RSAGNModelEnvParams":
        if self.output_globals_size > 0:
            power_logits = processed_graph.globals.reshape((-1,)) / self.temperature
        else:
            init_feature_array = jnp.zeros((params.k_paths, edge_features.shape[1]))
            # Define a body func to retrieve path slots and update action array
            def get_power_action_dist(i, feature_array):
                # Get the processed graph edge features corresponding to the i-th path
                path_features = get_path_slots(edge_features, params, nodes_sd, i, agg_func="sum").reshape((1, -1))
                # Update the array with the path features
                action_array = jax.lax.dynamic_update_slice(feature_array, path_features,(i, 0))
                return action_array
            path_feature_batch = jax.lax.fori_loop(0, params.k_paths, get_power_action_dist, init_feature_array)
            power_logits = path_mlp(path_feature_batch)
        if self.discrete:
            power_action_dist = distrax.Categorical(logits=power_logits)
        else:
            alpha = self.min_concentration + jax.nn.softplus(power_logits) * (
                    self.max_concentration - self.min_concentration
            )
            beta = self.min_concentration + jax.nn.softplus(power_logits) * (
                    self.max_concentration - self.min_concentration
            )
            power_action_dist = distrax.Beta(alpha, beta)

    # Return a distrax.Categorical distribution over actions (which can be masked later)
    path_action_dist = distrax.Categorical(logits=path_action_logits)
    return (path_action_dist, power_action_dist)

GraphNet

Bases: Module

A complete Graph Network model defined with Jraph.

Source code in xlron/models/models.py 
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
class GraphNet(nn.Module):
    """A complete Graph Network model defined with Jraph."""

    latent_size: int
    num_mlp_layers: int
    message_passing_steps: int
    output_globals_size: int = 0
    output_edges_size: int = 0
    output_nodes_size: int = 0
    dropout_rate: float = 0
    skip_connections: bool = True
    use_edge_model: bool = True
    gnn_layer_norm: bool = True
    mlp_layer_norm: bool = False
    deterministic: bool = True  # If true, no dropout (better for RL purposes)
    use_attention: bool = True

    @nn.compact
    def __call__(self, graphs: jraph.GraphsTuple) -> jraph.GraphsTuple:
        # Template code from here: https://github.com/google/flax/blob/main/examples/ogbg_molpcba/models.py
        # We will first linearly project the original features as 'embeddings'.
        embedder = jraph.GraphMapFeatures(
            embed_node_fn=nn.Dense(self.latent_size) if self.output_nodes_size > 0 else None,
            embed_edge_fn=nn.Dense(self.latent_size),
            embed_global_fn=nn.Dense(self.latent_size) if self.output_globals_size > 0 else None,
        )
        if graphs.edges.ndim >= 3:
            # Dims are (edges, slots, features e.g. power, source/dest)
            # Keep the leading dimension fixed and combine the remaining dimensions
            edges = graphs.edges.reshape((graphs.edges.shape[0], -1))
            graphs = graphs._replace(edges=edges)
        processed_graphs = embedder(graphs)

        # Now, we will apply a Graph Network once for each message-passing round.
        for _ in range(self.message_passing_steps):
            mlp_feature_sizes = [self.latent_size] * self.num_mlp_layers
            if self.use_edge_model:
                update_edge_fn = jraph.concatenated_args(
                    MLP(
                        mlp_feature_sizes,
                        dropout_rate=self.dropout_rate,
                        deterministic=self.deterministic,
                        layer_norm=self.mlp_layer_norm,
                    )
                )
            else:
                update_edge_fn = None

            update_node_fn = jraph.concatenated_args(
                MLP(
                    mlp_feature_sizes,
                    dropout_rate=self.dropout_rate,
                    deterministic=self.deterministic,
                    layer_norm=self.mlp_layer_norm,
                )
            )
            update_global_fn = jraph.concatenated_args(
                MLP(
                    mlp_feature_sizes,
                    dropout_rate=self.dropout_rate,
                    deterministic=self.deterministic,
                    layer_norm=self.mlp_layer_norm,
                )
            )

            if self.use_attention:
                def _attention_logit_fn(edges, sender_attr, receiver_attr, global_edge_attributes):
                    """Calculate attention logits using edges, nodes and global attributes."""
                    x = jnp.concatenate((edges, sender_attr, receiver_attr, global_edge_attributes), axis=1)
                    return MLP(mlp_feature_sizes, dropout_rate=self.dropout_rate,
                               deterministic=self.deterministic)(x)

                def _attention_reduce_fn(
                    edges: jnp.ndarray, attention: jnp.ndarray
                ) -> jnp.ndarray:
                    # TODO - try more sophisticated attention reduce function (not sure what it would be)
                    #  (here might help https://uvadlc-notebooks.readthedocs.io/en/latest/tutorial_notebooks/JAX/tutorial7/GNN_overview.html)
                    return attention * edges

                graph_net = GraphNetGAT(
                    update_node_fn=update_node_fn,
                    update_edge_fn=update_edge_fn,  # Update the edges with MLP prior to attention
                    update_global_fn=update_global_fn if self.output_globals_size > 0 else None,
                    attention_logit_fn=_attention_logit_fn,
                    attention_reduce_fn=_attention_reduce_fn,
                )
            else:
                graph_net = GraphNetwork(
                    update_node_fn=update_node_fn,
                    update_edge_fn=update_edge_fn,
                    update_global_fn=update_global_fn if self.output_globals_size > 0 else None,
                )

            if self.skip_connections:
                processed_graphs = add_graphs_tuples(
                graph_net(processed_graphs), processed_graphs
            )
            else:
                processed_graphs = graph_net(processed_graphs)

            if self.gnn_layer_norm:
                processed_graphs = processed_graphs._replace(
                    nodes=nn.LayerNorm()(processed_graphs.nodes),
                    edges=nn.LayerNorm()(processed_graphs.edges),
                    globals=nn.LayerNorm()(processed_graphs.globals) if processed_graphs.globals is not None else None,
                )

        decoder = jraph.GraphMapFeatures(
            embed_global_fn=nn.Dense(self.output_globals_size) if self.output_globals_size > 0 else None,
            embed_node_fn=nn.Dense(self.output_nodes_size) if self.output_nodes_size > 0 else None,
            embed_edge_fn=nn.Dense(self.output_edges_size),
        )
        processed_graphs = decoder(processed_graphs)

        return processed_graphs

LaunchPowerActorCriticMLP

Bases: Module

In this implementation, we take an observation of th current request + statistics on each of the K candidate paths. We make K forward passes, one for each path, and output a distribution over power levels for each path. In action selection, we then sample from each distribution and use the sampled power levels to mask paths for the routing heuristic, which then determines the path taken. The selected path index is then used to select which output action, distribution, and value to use for the loss calculation.

Source code in xlron/models/models.py 
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
class LaunchPowerActorCriticMLP(nn.Module):
    """In this implementation, we take an observation of th current request + statistics on each of the K candidate paths.
    We make K forward passes, one for each path, and output a distribution over power levels for each path.
    In action selection, we then sample from each distribution and use the sampled power levels to mask paths for the
    routing heuristic, which then determines the path taken. The selected path index is then used to select which output
    action, distribution, and value to use for the loss calculation."""
    action_dim: Sequence[int]
    activation: str = "tanh"
    num_layers: int = 2
    num_units: int = 64
    layer_norm: bool = False
    min_power_dbm: float = 0.0
    max_power_dbm: float = 2.0
    step_power_dbm: float = 0.1
    discrete: bool = True
    temperature: float = 1.0
    k_paths: int = 5
    num_base_features: int = 4
    num_path_features: int = 7
    # Beta distribution parameters (used only if discrete=False)
    min_concentration: float = 0.1
    max_concentration: float = 20.0
    epsilon: float = 1e-6

    @property
    def num_power_levels(self):
        """Calculate number of power levels dynamically"""
        return int((self.max_power_dbm - self.min_power_dbm) / self.step_power_dbm) + 1

    @property
    def power_levels(self):
        """Calculate power levels dynamically"""
        return jnp.linspace(self.min_power_dbm, self.max_power_dbm, self.num_power_levels)

    @nn.compact
    def __call__(self, x):
        # Helper function to create MLP layers
        def make_mlp(prefix):
            layers = []
            for i in range(self.num_layers):
                layers.append(nn.Dense(
                    self.num_units,
                    kernel_init=orthogonal(np.sqrt(2)),
                    name=f"{prefix}_dense_{i}"
                ))
                if self.layer_norm:
                    layers.append(nn.LayerNorm(name=f"{prefix}_norm_{i}"))
            return layers

        # Initialize actor network layers
        actor_net = make_mlp("actor")
        if self.discrete:
            actor_out = nn.Dense(
                self.num_power_levels,
                kernel_init=orthogonal(0.01),
                name="actor_output"
            )
        else:
            alpha_out = nn.Dense(1, kernel_init=orthogonal(0.01), name="alpha")
            beta_out = nn.Dense(1, kernel_init=orthogonal(0.01), name="beta")

        def activate(x):
            if self.activation == "relu": return jax.nn.relu(x)
            if self.activation == "crelu": return crelu(x)
            return jnp.tanh(x)

        def forward(x, layers):
            for layer in layers:
                x = layer(x)
                if isinstance(layer, nn.Dense):
                    x = activate(x)
            return x

        num_base_features = self.num_base_features
        num_path_features = self.num_path_features
        temperature = self.temperature
        discrete = self.discrete
        min_concentration = self.min_concentration
        max_concentration = self.max_concentration

        # Create a class to handle the scan
        class PathProcessor(nn.Module):
            def __call__(self, carry, i):
                base = x[:num_base_features]
                path = jax.lax.dynamic_slice(
                    x,
                    (num_base_features + i *num_path_features,),
                    (num_path_features,)
                )
                features = jnp.concatenate([base, path])
                actor_hidden = forward(features, actor_net)
                if discrete:
                    out = actor_out(actor_hidden) / temperature
                else:
                    alpha = min_concentration + jax.nn.softplus(alpha_out(actor_hidden)) * (
                            max_concentration - min_concentration
                    )
                    beta = min_concentration + jax.nn.softplus(beta_out(actor_hidden)) * (
                            max_concentration - min_concentration
                    )
                    out = (alpha, beta)

                return carry, out

        # Scan over paths
        _, dist_params = nn.scan(
            PathProcessor,
            variable_broadcast="params",
            split_rngs={"params": False},
            length=self.k_paths,
        )()(None, jnp.arange(self.k_paths))

        # Initialize critic network layers
        critic_net = make_mlp("critic")
        critic_out = nn.Dense(1, kernel_init=orthogonal(1.0), name="critic_output")
        value = jnp.squeeze(critic_out(forward(x, critic_net)), axis=-1)

        # Create appropriate distribution
        if self.discrete:
            dist = distrax.Categorical(logits=dist_params)
        else:
            dist = distrax.Beta(dist_params[0].reshape((self.k_paths,)), dist_params[1].reshape((self.k_paths,)))
        # N.B. that this is a single distribution object, but it is batched over K paths

        return [None, dist], value

    def sample_action(self, seed, dist, log_prob=False, deterministic=False):
        """Sample an action and convert to power level"""
        if self.discrete:
            if deterministic:
                # Take most probable action
                raw_action = dist.mode()
            else:
                # Sample from distribution
                raw_action = dist.sample(seed=seed)
            processed_action = self.power_levels[raw_action].reshape((self.k_paths, 1))
        else:
            if deterministic:
                # Use mean of Beta distribution for deterministic action
                mean = dist.alpha / (dist.alpha + dist.beta)
                raw_action = jnp.clip(mean, self.epsilon, 1.0 - self.epsilon)
            else:
                # Sample from Beta (clipping to avoid edge values with undefined gradient) and scale to power range
                raw_action = jnp.clip(dist.sample(seed=seed), self.epsilon, 1.0 - self.epsilon)
            processed_action = self.min_power_dbm + raw_action * (self.max_power_dbm - self.min_power_dbm)
        processed_action = from_dbm(processed_action)
        if log_prob:
            return processed_action, dist.log_prob(jnp.squeeze(raw_action))
        return processed_action

    def get_action_probs(self, dist):
        """Get probabilities for discrete case or pdf for continuous case"""
        if self.discrete:
            return dist.probs()
        else:
            x = jnp.linspace(0, 1, 100)
            return dist.prob(x)

num_power_levels property

Calculate number of power levels dynamically

power_levels property

Calculate power levels dynamically

get_action_probs(dist)

Get probabilities for discrete case or pdf for continuous case

Source code in xlron/models/models.py 
275
276
277
278
279
280
281
def get_action_probs(self, dist):
    """Get probabilities for discrete case or pdf for continuous case"""
    if self.discrete:
        return dist.probs()
    else:
        x = jnp.linspace(0, 1, 100)
        return dist.prob(x)

sample_action(seed, dist, log_prob=False, deterministic=False)

Sample an action and convert to power level

Source code in xlron/models/models.py 
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
def sample_action(self, seed, dist, log_prob=False, deterministic=False):
    """Sample an action and convert to power level"""
    if self.discrete:
        if deterministic:
            # Take most probable action
            raw_action = dist.mode()
        else:
            # Sample from distribution
            raw_action = dist.sample(seed=seed)
        processed_action = self.power_levels[raw_action].reshape((self.k_paths, 1))
    else:
        if deterministic:
            # Use mean of Beta distribution for deterministic action
            mean = dist.alpha / (dist.alpha + dist.beta)
            raw_action = jnp.clip(mean, self.epsilon, 1.0 - self.epsilon)
        else:
            # Sample from Beta (clipping to avoid edge values with undefined gradient) and scale to power range
            raw_action = jnp.clip(dist.sample(seed=seed), self.epsilon, 1.0 - self.epsilon)
        processed_action = self.min_power_dbm + raw_action * (self.max_power_dbm - self.min_power_dbm)
    processed_action = from_dbm(processed_action)
    if log_prob:
        return processed_action, dist.log_prob(jnp.squeeze(raw_action))
    return processed_action

MLP

Bases: Module

A multi-layer perceptron.

Source code in xlron/models/models.py 
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
class MLP(nn.Module):
    """A multi-layer perceptron."""

    feature_sizes: Sequence[int]
    dropout_rate: float = 0
    deterministic: bool = True
    activation: Callable[[jnp.ndarray], jnp.ndarray] = nn.tanh
    layer_norm: bool = False

    @nn.compact
    def __call__(self, inputs):
        x = inputs
        for size in self.feature_sizes:
            x = nn.Dense(features=size)(x)
            x = self.activation(x)
            x = nn.Dropout(rate=self.dropout_rate, deterministic=self.deterministic)(
                x
            )
            if self.layer_norm:
                x = nn.LayerNorm()(x)
        return x

add_graphs_tuples(graphs, other_graphs)

Adds the nodes, edges and global features from other_graphs to graphs.

Source code in xlron/models/models.py 
30
31
32
33
34
35
36
37
38
def add_graphs_tuples(
    graphs: jraph.GraphsTuple, other_graphs: jraph.GraphsTuple
) -> jraph.GraphsTuple:
    """Adds the nodes, edges and global features from other_graphs to graphs."""
    return graphs._replace(
        nodes=graphs.nodes + other_graphs.nodes,
        edges=graphs.edges + other_graphs.edges,
        globals=graphs.globals + other_graphs.globals if graphs.globals is not None else None,
    )

crelu(x)

Computes the Concatenated ReLU (CReLU) activation function.

Source code in xlron/models/models.py 
24
25
26
27
def crelu(x):
    """Computes the Concatenated ReLU (CReLU) activation function."""
    x = jnp.concatenate([x, -x], axis=-1)
    return nn.relu(x)

Heuristics

DeepRMSAEnvState

Bases: RSAEnvState

Dataclass to hold environment state for DeepRMSA.

Parameters:

Name Type Description Default
path_stats Array

Path stats array containing

 required
Source code in xlron/environments/dataclasses.py 
191
192
193
194
195
196
197
198
199
200
201
202
@struct.dataclass
class DeepRMSAEnvState(RSAEnvState):
    """Dataclass to hold environment state for DeepRMSA.

    Args:
        path_stats (chex.Array): Path stats array containing
        1. Required slots on path
        2. Total available slots on path
        3. Size of 1st free spectrum block
        4. Avg. free block size
    """
    path_stats: chex.Array

EnvParams

Dataclass to hold environment parameters. Parameters are immutable.

Parameters:

Name Type Description Default
max_requests Scalar

Maximum number of requests in an episode

 required
incremental_loading Scalar

Incremental increase in traffic load (non-expiring requests)

 required
end_first_blocking Scalar

End episode on first blocking event

 required
continuous_operation Scalar

If True, do not reset the environment at the end of an episode

 required
edges Array

Two column array defining source-dest node-pair edges of the graph

 required
slot_size Scalar

Spectral width of frequency slot in GHz

 required
consider_modulation_format Scalar

If True, consider modulation format to determine required slots

 required
link_length_array Array

Array of link lengths

 required
aggregate_slots Scalar

Number of slots to aggregate into a single action (First-Fit with aggregation)

 required
guardband Scalar

Guard band in slots

 required
directed_graph bool

Whether graph is directed (one fibre per link per transmission direction)

 required
Source code in xlron/environments/dataclasses.py 
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
@struct.dataclass
class EnvParams:
    """Dataclass to hold environment parameters. Parameters are immutable.

    Args:
        max_requests (chex.Scalar): Maximum number of requests in an episode
        incremental_loading (chex.Scalar): Incremental increase in traffic load (non-expiring requests)
        end_first_blocking (chex.Scalar): End episode on first blocking event
        continuous_operation (chex.Scalar): If True, do not reset the environment at the end of an episode
        edges (chex.Array): Two column array defining source-dest node-pair edges of the graph
        slot_size (chex.Scalar): Spectral width of frequency slot in GHz
        consider_modulation_format (chex.Scalar): If True, consider modulation format to determine required slots
        link_length_array (chex.Array): Array of link lengths
        aggregate_slots (chex.Scalar): Number of slots to aggregate into a single action (First-Fit with aggregation)
        guardband (chex.Scalar): Guard band in slots
        directed_graph (bool): Whether graph is directed (one fibre per link per transmission direction)
    """
    max_requests: chex.Scalar = struct.field(pytree_node=False)
    incremental_loading: chex.Scalar = struct.field(pytree_node=False)
    end_first_blocking: chex.Scalar = struct.field(pytree_node=False)
    continuous_operation: chex.Scalar = struct.field(pytree_node=False)
    edges: chex.Array = struct.field(pytree_node=False)
    slot_size: chex.Scalar = struct.field(pytree_node=False)
    consider_modulation_format: chex.Scalar = struct.field(pytree_node=False)
    link_length_array: chex.Array = struct.field(pytree_node=False)
    aggregate_slots: chex.Scalar = struct.field(pytree_node=False)
    guardband: chex.Scalar = struct.field(pytree_node=False)
    directed_graph: bool = struct.field(pytree_node=False)
    maximise_throughput: bool = struct.field(pytree_node=False)
    reward_type: str = struct.field(pytree_node=False)
    values_bw: chex.Array = struct.field(pytree_node=False)
    truncate_holding_time: bool = struct.field(pytree_node=False)
    traffic_array: bool = struct.field(pytree_node=False)

EnvState

Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

Parameters:

Name Type Description Default
current_time Scalar

Current time in environment

 required
holding_time Scalar

Holding time of current request

 required
total_timesteps Scalar

Total timesteps in environment

 required
total_requests Scalar

Total requests in environment

 required
graph GraphsTuple

Graph tuple representing network state

 required
full_link_slot_mask Array

Action mask for link slot action (including if slot actions are aggregated)

 required
accepted_services Array

Number of accepted services

 required
accepted_bitrate Array

Accepted bitrate

 required
Source code in xlron/environments/dataclasses.py 
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
@struct.dataclass
class EnvState:
    """Dataclass to hold environment state. State is mutable and arrays are traced on JIT compilation.

    Args:
        current_time (chex.Scalar): Current time in environment
        holding_time (chex.Scalar): Holding time of current request
        total_timesteps (chex.Scalar): Total timesteps in environment
        total_requests (chex.Scalar): Total requests in environment
        graph (jraph.GraphsTuple): Graph tuple representing network state
        full_link_slot_mask (chex.Array): Action mask for link slot action (including if slot actions are aggregated)
        accepted_services (chex.Array): Number of accepted services
        accepted_bitrate (chex.Array): Accepted bitrate
        """
    current_time: chex.Scalar
    holding_time: chex.Scalar
    total_timesteps: chex.Scalar
    total_requests: chex.Scalar
    graph: jraph.GraphsTuple
    full_link_slot_mask: chex.Array
    accepted_services: chex.Array
    accepted_bitrate: chex.Array
    total_bitrate: chex.Array
    list_of_requests: chex.Array

GNModelEnvParams

Bases: RSAEnvParams

Dataclass to hold environment state for GN model environments.

Source code in xlron/environments/dataclasses.py 
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
@struct.dataclass
class GNModelEnvParams(RSAEnvParams):
    """Dataclass to hold environment state for GN model environments.
    """
    ref_lambda: chex.Scalar = struct.field(pytree_node=False)
    max_spans: chex.Scalar = struct.field(pytree_node=False)
    max_span_length: chex.Scalar = struct.field(pytree_node=False)
    nonlinear_coeff: chex.Scalar = struct.field(pytree_node=False)
    raman_gain_slope: chex.Scalar = struct.field(pytree_node=False)
    attenuation: chex.Scalar = struct.field(pytree_node=False)
    attenuation_bar: chex.Scalar = struct.field(pytree_node=False)
    dispersion_coeff: chex.Scalar = struct.field(pytree_node=False)
    dispersion_slope: chex.Scalar = struct.field(pytree_node=False)
    noise_figure: chex.Scalar = struct.field(pytree_node=False)
    coherent: bool = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    num_roadms: chex.Scalar = struct.field(pytree_node=False)
    roadm_loss: chex.Scalar = struct.field(pytree_node=False)
    num_spans: chex.Scalar = struct.field(pytree_node=False)
    launch_power_type: chex.Scalar = struct.field(pytree_node=False)
    snr_margin: chex.Scalar = struct.field(pytree_node=False)
    max_snr: chex.Scalar = struct.field(pytree_node=False)
    max_power: chex.Scalar = struct.field(pytree_node=False)
    min_power: chex.Scalar = struct.field(pytree_node=False)
    step_power: chex.Scalar = struct.field(pytree_node=False)
    last_fit: bool = struct.field(pytree_node=False)
    default_launch_power: chex.Scalar = struct.field(pytree_node=False)
    mod_format_correction: bool = struct.field(pytree_node=False)

GNModelEnvState

Bases: RSAEnvState

Dataclass to hold environment state for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
276
277
278
279
280
281
282
283
284
285
286
287
@struct.dataclass
class GNModelEnvState(RSAEnvState):
    """Dataclass to hold environment state for RSA with GN model.
    """
    link_snr_array: chex.Array  # Available SNR on each link
    channel_centre_bw_array: chex.Array  # Channel centre bandwidth for each active connection
    path_index_array: chex.Array  # Contains indices of lightpaths in use on slots (used for lightpath SNR calculation)
    channel_power_array: chex.Array  # Channel power for each active connection
    channel_centre_bw_array_prev: chex.Array  # Channel centre bandwidth for each active connection in previous timestep
    path_index_array_prev: chex.Array  # Contains indices of lightpaths in use on slots in previous timestep
    channel_power_array_prev: chex.Array  # Channel power for each active connection in previous timestep
    launch_power_array: chex.Array  # Launch power array

LogEnvState

Dataclass to hold environment state for logging.

Parameters:

Name Type Description Default
env_state EnvState

Environment state

 required
lengths Scalar

Lengths

 required
returns Scalar

Returns

 required
cum_returns Scalar

Cumulative returns

 required
episode_lengths Scalar

Episode lengths

 required
episode_returns Scalar

Episode returns

 required
accepted_services Scalar

Accepted services

 required
accepted_bitrate Scalar

Accepted bitrate

 required
done Scalar

Done

 required
Source code in xlron/environments/dataclasses.py 
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
@struct.dataclass
class LogEnvState:
    """Dataclass to hold environment state for logging.

    Args:
        env_state (EnvState): Environment state
        lengths (chex.Scalar): Lengths
        returns (chex.Scalar): Returns
        cum_returns (chex.Scalar): Cumulative returns
        episode_lengths (chex.Scalar): Episode lengths
        episode_returns (chex.Scalar): Episode returns
        accepted_services (chex.Scalar): Accepted services
        accepted_bitrate (chex.Scalar): Accepted bitrate
        done (chex.Scalar): Done
    """
    env_state: EnvState
    lengths: float
    returns: float
    cum_returns: float
    accepted_services: int
    accepted_bitrate: float
    total_bitrate: float
    utilisation: float
    done: bool

MultiBandRSAEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
237
238
239
240
241
242
@struct.dataclass
class MultiBandRSAEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for MultiBandRSA (RBSA).
    """
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)

MultiBandRSAEnvState

Bases: RSAEnvState

Dataclass to hold environment state for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
230
231
232
233
234
@struct.dataclass
class MultiBandRSAEnvState(RSAEnvState):
    """Dataclass to hold environment state for MultiBandRSA (RBSA).
    """
    pass

RMSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
306
307
308
309
310
311
312
313
@struct.dataclass
class RMSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulations_array: chex.Array = struct.field(pytree_node=False)

RMSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RMSA with GN model.

Parameters:

Name Type Description Default
link_snr_array Array

Link SNR array

 required
Source code in xlron/environments/dataclasses.py 
316
317
318
319
320
321
322
323
324
325
@struct.dataclass
class RMSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RMSA with GN model.

    Args:
        link_snr_array (chex.Array): Link SNR array
    """
    modulation_format_index_array: chex.Array  # Modulation format index for each active connection
    modulation_format_index_array_prev: chex.Array  # Modulation format index for each active connection in previous timestep
    mod_format_mask: chex.Array  # Modulation format mask

RSAEnvParams

Bases: EnvParams

Dataclass to hold environment parameters for RSA.

Parameters:

Name Type Description Default
num_nodes Scalar

Number of nodes

 required
num_links Scalar

Number of links

 required
link_resources Scalar

Number of link resources

 required
k_paths Scalar

Number of paths

 required
mean_service_holding_time Scalar

Mean service holding time

 required
load Scalar

Load

 required
arrival_rate Scalar

Arrival rate

 required
path_link_array Array

Path link array

 required
random_traffic bool

Random traffic matrix for RSA on each reset (else uniform or custom)

 required
max_slots Scalar

Maximum number of slots

 required
path_se_array Array

Path spectral efficiency array

 required
deterministic_requests bool

If True, use deterministic requests

 required
multiple_topologies bool

If True, use multiple topologies

 required
Source code in xlron/environments/dataclasses.py 
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
@struct.dataclass
class RSAEnvParams(EnvParams):
    """Dataclass to hold environment parameters for RSA.

    Args:
        num_nodes (chex.Scalar): Number of nodes
        num_links (chex.Scalar): Number of links
        link_resources (chex.Scalar): Number of link resources
        k_paths (chex.Scalar): Number of paths
        mean_service_holding_time (chex.Scalar): Mean service holding time
        load (chex.Scalar): Load
        arrival_rate (chex.Scalar): Arrival rate
        path_link_array (chex.Array): Path link array
        random_traffic (bool): Random traffic matrix for RSA on each reset (else uniform or custom)
        max_slots (chex.Scalar): Maximum number of slots
        path_se_array (chex.Array): Path spectral efficiency array
        deterministic_requests (bool): If True, use deterministic requests
        multiple_topologies (bool): If True, use multiple topologies
    """
    num_nodes: chex.Scalar = struct.field(pytree_node=False)
    num_links: chex.Scalar = struct.field(pytree_node=False)
    link_resources: chex.Scalar = struct.field(pytree_node=False)
    k_paths: chex.Scalar = struct.field(pytree_node=False)
    mean_service_holding_time: chex.Scalar = struct.field(pytree_node=False)
    load: chex.Scalar = struct.field(pytree_node=False)
    arrival_rate: chex.Scalar = struct.field(pytree_node=False)
    path_link_array: chex.Array = struct.field(pytree_node=False)
    random_traffic: bool = struct.field(pytree_node=False)
    max_slots: chex.Scalar = struct.field(pytree_node=False)
    path_se_array: chex.Array = struct.field(pytree_node=False)
    deterministic_requests: bool = struct.field(pytree_node=False)
    multiple_topologies: bool = struct.field(pytree_node=False)
    log_actions: bool = struct.field(pytree_node=False)
    disable_node_features: bool = struct.field(pytree_node=False)

RSAEnvState

Bases: EnvState

Dataclass to hold environment state for RSA.

Parameters:

Name Type Description Default
link_slot_array Array

Link slot array

 required
request_array Array

Request array

 required
link_slot_departure_array Array

Link slot departure array

 required
link_slot_mask Array

Link slot mask

 required
traffic_matrix Array

Traffic matrix

 required
Source code in xlron/environments/dataclasses.py 
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
@struct.dataclass
class RSAEnvState(EnvState):
    """Dataclass to hold environment state for RSA.

    Args:
        link_slot_array (chex.Array): Link slot array
        request_array (chex.Array): Request array
        link_slot_departure_array (chex.Array): Link slot departure array
        link_slot_mask (chex.Array): Link slot mask
        traffic_matrix (chex.Array): Traffic matrix
    """
    link_slot_array: chex.Array
    request_array: chex.Array
    link_slot_departure_array: chex.Array
    link_slot_mask: chex.Array
    traffic_matrix: chex.Array

RSAGNModelEnvParams

Bases: GNModelEnvParams

Dataclass to hold environment params for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
290
291
292
293
294
@struct.dataclass
class RSAGNModelEnvParams(GNModelEnvParams):
    """Dataclass to hold environment params for RSA with GN model.
    """
    pass

RSAGNModelEnvState

Bases: GNModelEnvState

Dataclass to hold environment state for RSA with GN model.

Source code in xlron/environments/dataclasses.py 
297
298
299
300
301
302
303
@struct.dataclass
class RSAGNModelEnvState(GNModelEnvState):
    """Dataclass to hold environment state for RSA with GN model.
    """
    active_lightpaths_array: chex.Array  # Active lightpath array. 1 x M array. Each value is a lightpath index. Used to calculate total throughput.
    active_lightpaths_array_departure: chex.Array  # Active lightpath array departure time.
    current_throughput: chex.Array  # Current throughput

RSAMultibandEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
335
336
337
338
339
340
@struct.dataclass
class RSAMultibandEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for MultiBandRSA (RBSA).
    """
    gap_start: chex.Scalar = struct.field(pytree_node=False)
    gap_width: chex.Scalar = struct.field(pytree_node=False)

RSAMultibandEnvState

Bases: RSAEnvState

Dataclass to hold environment state for MultiBandRSA (RBSA).

Source code in xlron/environments/dataclasses.py 
328
329
330
331
332
@struct.dataclass
class RSAMultibandEnvState(RSAEnvState):
    """Dataclass to hold environment state for MultiBandRSA (RBSA).
    """
    pass

RWALightpathReuseEnvState

Bases: RSAEnvState

Dataclass to hold environment state for RWA with lightpath reuse.

Parameters:

Name Type Description Default
path_index_array Array

Contains indices of lightpaths in use on slots

 required
path_capacity_array Array

Contains remaining capacity of each lightpath

 required
link_capacity_array Array

Contains remaining capacity of lightpath on each link-slot

 required
Source code in xlron/environments/dataclasses.py 
210
211
212
213
214
215
216
217
218
219
220
221
222
@struct.dataclass
class RWALightpathReuseEnvState(RSAEnvState):
    """Dataclass to hold environment state for RWA with lightpath reuse.

    Args:
        path_index_array (chex.Array): Contains indices of lightpaths in use on slots
        path_capacity_array (chex.Array): Contains remaining capacity of each lightpath
        link_capacity_array (chex.Array): Contains remaining capacity of lightpath on each link-slot
    """
    path_index_array: chex.Array  # Contains indices of lightpaths in use on slots
    path_capacity_array: chex.Array  # Contains remaining capacity of each lightpath
    link_capacity_array: chex.Array  # Contains remaining capacity of lightpath on each link-slot
    time_since_last_departure: chex.Array  # Time since last departure

VONEEnvParams

Bases: RSAEnvParams

Dataclass to hold environment parameters for VONE.

Parameters:

Name Type Description Default
node_resources Scalar

Number of node resources

 required
max_edges Scalar

Maximum number of edges

 required
min_node_resources Scalar

Minimum number of node resources

 required
max_node_resources Scalar

Maximum number of node resources

 required
Source code in xlron/environments/dataclasses.py 
369
370
371
372
373
374
375
376
377
378
379
380
381
382
@struct.dataclass
class VONEEnvParams(RSAEnvParams):
    """Dataclass to hold environment parameters for VONE.

    Args:
        node_resources (chex.Scalar): Number of node resources
        max_edges (chex.Scalar): Maximum number of edges
        min_node_resources (chex.Scalar): Minimum number of node resources
        max_node_resources (chex.Scalar): Maximum number of node resources
    """
    node_resources: chex.Scalar = struct.field(pytree_node=False)
    max_edges: chex.Scalar = struct.field(pytree_node=False)
    min_node_resources: chex.Scalar = struct.field(pytree_node=False)
    max_node_resources: chex.Scalar = struct.field(pytree_node=False)

VONEEnvState

Bases: RSAEnvState

Dataclass to hold environment state for VONE.

Parameters:

Name Type Description Default
node_capacity_array Array

Node capacity array

 required
node_resource_array Array

Node resource array

 required
node_departure_array Array

Node departure array

 required
action_counter Array

Action counter

 required
action_history Array

Action history

 required
node_mask_s Array

Node mask for source node

 required
node_mask_d Array

Node mask for destination node

 required
virtual_topology_patterns Array

Virtual topology patterns

 required
values_nodes Array

Values for nodes

 required
Source code in xlron/environments/dataclasses.py 
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
@struct.dataclass
class VONEEnvState(RSAEnvState):
    """Dataclass to hold environment state for VONE.

    Args:
        node_capacity_array (chex.Array): Node capacity array
        node_resource_array (chex.Array): Node resource array
        node_departure_array (chex.Array): Node departure array
        action_counter (chex.Array): Action counter
        action_history (chex.Array): Action history
        node_mask_s (chex.Array): Node mask for source node
        node_mask_d (chex.Array): Node mask for destination node
        virtual_topology_patterns (chex.Array): Virtual topology patterns
        values_nodes (chex.Array): Values for nodes
    """
    node_capacity_array: chex.Array
    node_resource_array: chex.Array
    node_departure_array: chex.Array
    action_counter: chex.Array
    action_history: chex.Array
    node_mask_s: chex.Array
    node_mask_d: chex.Array
    virtual_topology_patterns: chex.Array
    values_nodes: chex.Array

aggregate_slots(full_mask, params)

Aggregate slot mask to reduce action space. Only used if the --aggregate_slots flag is set to > 1. Aggregated action is valid if there is one valid slot action within the aggregated action window.

Parameters:

Name Type Description Default
full_mask Array

slot mask

 required
params EnvParams

environment parameters

 required

Returns:

Name Type Description
agg_mask Array

aggregated slot mask
Source code in xlron/environments/env_funcs.py 
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
@partial(jax.jit, static_argnums=(1,))
def aggregate_slots(full_mask: chex.Array, params: EnvParams) -> chex.Array:
    """Aggregate slot mask to reduce action space. Only used if the --aggregate_slots flag is set to > 1.
    Aggregated action is valid if there is one valid slot action within the aggregated action window.

    Args:
        full_mask: slot mask
        params: environment parameters

    Returns:
        agg_mask: aggregated slot mask
    """

    num_actions = math.ceil(params.link_resources/params.aggregate_slots)
    agg_mask = jnp.zeros((params.k_paths, num_actions), dtype=jnp.float32)

    def get_max(i, mask_val):
        """Get maximum value of array slice of length aggregate_slots."""
        mask_slice = jax.lax.dynamic_slice(
                mask_val,
                (0, i * params.aggregate_slots,),
                (1,  params.aggregate_slots,),
            )
        max_slice = jnp.max(mask_slice).reshape(1, -1)
        return max_slice

    def update_window_max(i, val):
        """Update ith index 'agg_mask' with max of ith slice of length aggregate_slots from 'full_mask'.

        Args:
            i: increments as += aggregate_slots
            val: tuple of (agg_mask, path_mask, path_index).
        Returns:
            new_agg_mask: agg_mask is updated with max of path_mask for window size aggregate_slots
            mask: mask is unchanged
            path_index: path_index is unchanged
        """
        agg_mask = val[0]
        full_mask = val[1]
        path_index = val[2]
        new_agg_mask = jax.lax.dynamic_update_slice(
            agg_mask,
            get_max(i, full_mask),
            (path_index, i),
        )
        return new_agg_mask, full_mask, path_index

    def apply_to_path_mask(i, val):
        """
        Loop through each path for num_actions steps and get_window_max at each step.

        Args:
            i: path index
            val: tuple of (agg_mask, mask) where mask is original link-slot mask and agg_mask is resulting aggregated mask
        Returns:
            new_agg_mask: agg_mask is updated with aggregated path mask
            mask: mask is unchanged
        """
        val = (
            val[0],  # aggregated mask (to be updated)
            val[1][i].reshape(1, -1),  # mask for path i
            i  # path index
        )
        new_agg_mask = jax.lax.fori_loop(
            0,
            num_actions,
            update_window_max,
            val,
        )[0]
        return new_agg_mask, full_mask

    return jax.lax.fori_loop(
            0,
            params.k_paths,
            apply_to_path_mask,
            (agg_mask, full_mask),
        )

best_fit(state, params)

Best-Fit Spectrum Allocation. Returns the best fit slot for each path.

Source code in xlron/heuristics/heuristics.py 
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
def best_fit(state: EnvState, params: EnvParams) -> chex.Array:
    """Best-Fit Spectrum Allocation. Returns the best fit slot for each path."""
    mask = get_action_mask(state, params)
    link_slot_array = jnp.where(state.link_slot_array < 0, 1., state.link_slot_array)
    nodes_sd, requested_bw = read_rsa_request(state.request_array)

    # We need to define a wrapper function in order to vmap with named arguments
    def _find_block_sizes(arr, starts_only=False, reverse=True):
        return jax.vmap(find_block_sizes, in_axes=(0, None, None))(arr, starts_only, reverse)

    block_sizes_right = _find_block_sizes(link_slot_array, starts_only=False, reverse=False)
    block_sizes_left = _find_block_sizes(link_slot_array, starts_only=False, reverse=True)
    block_sizes = jnp.maximum((block_sizes_left + block_sizes_right) - 1, 0)
    paths = get_paths(params, nodes_sd)
    se = get_paths_se(params, nodes_sd) if params.consider_modulation_format else jnp.ones((params.k_paths,))
    num_slots = jax.vmap(required_slots, in_axes=(None, 0, None, None))(requested_bw, se, params.slot_size, params.guardband)

    # Quantify how well the request fits within a free spectral block
    def get_bf_on_path(path, blocks, req_slots):
        fits = jax.vmap(lambda x: x - req_slots, in_axes=0)(blocks)
        fits = jnp.where(fits >= 0, fits, params.link_resources)
        path_fit = jnp.dot(path, fits) / jnp.sum(path)
        return path_fit
    fits_block = jax.vmap(lambda x, y, z: get_bf_on_path(x, y, z), in_axes=(0, None, 0))(paths, block_sizes, num_slots)

    # Quantity much of a gap there is between the assigned slots and the next occupied slots on the left
    def get_bf_on_path_left(path, blocks, req_slots):
        fits = jax.vmap(lambda x: x - req_slots, in_axes=0)(blocks)
        fits = jnp.where(fits >= 0, fits, params.link_resources)
        fits_shift = jax.lax.dynamic_slice(fits, (0, 1), (fits.shape[0], fits.shape[1]-1))
        fits_shift = jnp.concatenate((jnp.full((fits.shape[0], 1), params.link_resources), fits_shift), axis=1)
        fits = fits + 1/jnp.maximum(fits_shift, 1)
        path_fit = jnp.dot(path, fits) / jnp.sum(path)
        return path_fit
    fits_left = jax.vmap(lambda x, y, z: get_bf_on_path_left(x, y, z), in_axes=(0, None, 0))(paths, block_sizes_left, num_slots)

    # Quantity much of a gap there is between the assigned slots and the next occupied slots on the right
    def get_bf_on_path_right(path, blocks, req_slots):
        fits = jax.vmap(lambda x: x - req_slots, in_axes=0)(blocks)
        fits = jnp.where(fits >= 0, fits, params.link_resources)
        fits_shift = jax.lax.dynamic_slice(fits, (0, 0), (fits.shape[0], fits.shape[1] - 1))
        fits_shift = jnp.concatenate((fits_shift, jnp.full((fits.shape[0], 1), params.link_resources)), axis=1)
        fits = fits + 1/jnp.maximum(fits_shift, 1)
        path_fit = jnp.dot(path, fits) / jnp.sum(path)
        return path_fit
    fits_right = jax.vmap(lambda x, y, z: get_bf_on_path_right(x, y, z), in_axes=(0, None, 0))(paths, block_sizes_right, num_slots)

    # Sum the contribution to the overall quality of fit, and scale down the left/right contributions
    fits = jnp.sum(jnp.stack((fits_block, fits_left/params.link_resources, fits_right/params.link_resources), axis=0), axis=0)
    # Mask out occupied lightpaths (in case the quality of fit on some links is good enough to be considered, even if the overall path is invalid)
    fits = jnp.where(mask == 0, jnp.inf, fits)
    best_slots = jnp.argmin(fits, axis=1)
    best_fits = jnp.min(fits, axis=1)
    return best_slots, best_fits

bf_ksp(state, params)

Get the first available slot from the first k-shortest paths Method: Go through action mask and find the first available slot on all paths

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required

Returns:

Type Description
Array

chex.Array: Action
Source code in xlron/heuristics/heuristics.py 
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
@partial(jax.jit, static_argnums=(1,))
def bf_ksp(state: EnvState, params: EnvParams) -> chex.Array:
    """Get the first available slot from the first k-shortest paths
    Method: Go through action mask and find the first available slot on all paths

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters

    Returns:
        chex.Array: Action
    """
    best_slots, fitness = best_fit(state, params)
    # Chosen path is the one with the best fit
    path_index = jnp.argmin(fitness)
    slot_index = best_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

calculate_path_capacity(path_length, min_request=100, scale_factor=1.0, alpha=0.0002, NF=4.5, B=10000000000000.0, R_s=100000000000.0, beta_2=-2.17e-26, gamma=0.0012, L_s=100000.0, lambda0=1.55e-06)

From Nevin JOCN paper 2022: https://discovery.ucl.ac.uk/id/eprint/10175456/1/RL_JOCN_accepted.pdf

Source code in xlron/environments/env_funcs.py 
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
def calculate_path_capacity(
        path_length,
        min_request=100,  # Minimum data rate request size
        scale_factor=1.0,  # Scale factor for link capacity
        alpha=0.2e-3, # Fibre attenuation coefficient
        NF=4.5,  # Amplifier noise figure
        B=10e12,  # Total modulated bandwidth
        R_s=100e9,  # Symbol rate
        beta_2=-21.7e-27,  # Dispersion parameter
        gamma=1.2e-3,  # Nonlinear coefficient
        L_s=100e3,  # Span length
        lambda0=1550e-9,  # Wavelength
):
    """From Nevin JOCN paper 2022: https://discovery.ucl.ac.uk/id/eprint/10175456/1/RL_JOCN_accepted.pdf"""
    alpha_lin = alpha / 4.343  # Linear attenuation coefficient
    N_spans = jnp.floor(path_length * 1e3 / L_s)  # Number of fibre spans on path
    L_eff = (1 - jnp.exp(-alpha_lin * L_s)) / alpha_lin  # Effective length of span in m
    sigma_2_ase = (jnp.exp(alpha_lin * L_s) - 1) * 10**(NF/10) * 6.626e-34 * 2.998e8 * R_s / lambda0  # ASE noise power
    span_NSR = jnp.cbrt(2 * sigma_2_ase**2 * alpha_lin * gamma**2 * L_eff**2 *
                        jnp.log(jnp.pi**2 * jnp.abs(beta_2) * B**2 / alpha_lin) / (jnp.pi * jnp.abs(beta_2) * R_s**2))  # Noise-to-signal ratio per span
    path_NSR = jnp.where(N_spans < 1, 1, N_spans) * span_NSR  # Noise-to-signal ratio per path
    path_capacity = 2 * R_s/1e9 * jnp.log2(1 + 1/path_NSR)  # Capacity of path in Gbps
    # Round link capacity down to nearest increment of minimum request size and apply scale factor
    path_capacity = jnp.floor(path_capacity * scale_factor / min_request) * min_request
    return path_capacity

calculate_path_stats(state, params, request)

For use in DeepRMSA agent observation space. Calculate: 1. Size of 1st suitable free spectrum block 2. Index of 1st suitable free spectrum block 3. Required slots on path 4. Avg. free block size 5. Free slots

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required
request Array

Request array in format [source_node, data-rate, destination_node]

 required

Returns:

Name Type Description
stats Array

Array of calculated path statistics
Source code in xlron/environments/env_funcs.py 
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
@partial(jax.jit, static_argnums=(1,))
def calculate_path_stats(state: EnvState, params: EnvParams, request: chex.Array) -> chex.Array:
    """For use in DeepRMSA agent observation space.
    Calculate:
    1. Size of 1st suitable free spectrum block
    2. Index of 1st suitable free spectrum block
    3. Required slots on path
    4. Avg. free block size
    5. Free slots

    Args:
        state: Environment state
        params: Environment parameters
        request: Request array in format [source_node, data-rate, destination_node]

    Returns:
        stats: Array of calculated path statistics
    """
    nodes_sd, requested_datarate = read_rsa_request(request)
    init_val = jnp.zeros((params.k_paths, 5))
    # TODO - check if the normalisation is useful
    def body_fun(i, val):
        slots = get_path_slots(state.link_slot_array, params, nodes_sd, i)
        se = get_paths_se(params, nodes_sd)[i] if params.consider_modulation_format else jnp.array([1])
        req_slots = jnp.squeeze(required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband))
        req_slots_norm = req_slots#*params.slot_size / jnp.max(params.values_bw.val)
        free_slots_norm = jnp.sum(jnp.where(slots == 0, 1, 0)) #/ params.link_resources
        block_sizes = find_block_sizes(slots)
        first_block_index = jnp.argmax(block_sizes >= req_slots)
        first_block_index_norm = first_block_index #/ params.link_resources
        first_block_size_norm = jnp.squeeze(
            jax.lax.dynamic_slice(block_sizes, (first_block_index,), (1,))
        ) / req_slots
        avg_block_size_norm = jnp.sum(block_sizes) / jnp.max(jnp.array([jnp.sum(find_block_starts(slots)), 1])) #/ req_slots
        val = jax.lax.dynamic_update_slice(
            val,
            jnp.array([[first_block_size_norm, first_block_index_norm, req_slots_norm, avg_block_size_norm, free_slots_norm]]),
            (i, 0)
        )  # N.B. that all values are normalised
        return val

    stats = jax.lax.fori_loop(
            0,
            params.k_paths,
            body_fun,
            init_val,
        )

    return stats

check_action_rmsa_gn_model(state, action, params)

Check if action is valid for RSA GN model Args: state (EnvState): Environment state params (EnvParams): Environment parameters action (chex.Array): Action array Returns: bool: True if action is invalid, False if action is valid

Source code in xlron/environments/env_funcs.py 
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
@partial(jax.jit, static_argnums=(1,))
def check_action_rmsa_gn_model(state: EnvState, action: Optional[chex.Array], params: EnvParams) -> bool:
    """Check if action is valid for RSA GN model
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
        action (chex.Array): Action array
    Returns:
        bool: True if action is invalid, False if action is valid
    """
    # Check if action is valid
    # TODO - log failure reasons in info
    snr_sufficient_check = check_snr_sufficient(state, params)
    spectrum_reuse_check = check_no_spectrum_reuse(state.link_slot_array)
    # jax.debug.print("spectrum_reuse_check {}", spectrum_reuse_check, ordered=True)
    # jax.debug.print("snr_sufficient_check {}", snr_sufficient_check, ordered=True)
    return jnp.any(jnp.stack((
        spectrum_reuse_check,
        snr_sufficient_check,
    )))

check_action_rsa(state)

Check if action is valid. Combines checks for: - no spectrum reuse

Parameters:

Name Type Description Default
state

current state

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
def check_action_rsa(state):
    """Check if action is valid.
    Combines checks for:
    - no spectrum reuse

    Args:
        state: current state

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.any(jnp.stack((
        check_no_spectrum_reuse(state.link_slot_array),
    )))

check_action_rwalr(state, action, params)

Combines checks for: - no spectrum reuse - lightpath available and existing

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Name Type Description
bool bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
def check_action_rwalr(state: EnvState, action: chex.Array, params: EnvParams) -> bool:
    """Combines checks for:
    - no spectrum reuse
    - lightpath available and existing

    Args:
        state: Environment state

    Returns:
        bool: True if check failed, False if check passed

    """
    return jnp.any(jnp.stack((
        check_no_spectrum_reuse(state.link_slot_array),
        jnp.logical_not(check_lightpath_available_and_existing(state, params, action)[0]),
    )))

check_all_nodes_assigned(node_departure_array, total_requested_nodes)

Count negative values on each node (row) in node departure array, sum them, must equal total requested_nodes.

Parameters:

Name Type Description Default
node_departure_array Array

Node departure array (N x R) where N is number of nodes and R is number of resources

 required
total_requested_nodes int

Total requested nodes (int)

 required

Returns:

Name Type Description
bool bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
def check_all_nodes_assigned(node_departure_array: chex.Array, total_requested_nodes: int) -> bool:
    """Count negative values on each node (row) in node departure array, sum them, must equal total requested_nodes.

    Args:
        node_departure_array: Node departure array (N x R) where N is number of nodes and R is number of resources
        total_requested_nodes: Total requested nodes (int)

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.sum(jnp.sum(jnp.where(node_departure_array < 0, 1, 0), axis=1)) != total_requested_nodes

check_lightpath_available_and_existing(state, params, action)

Check if lightpath is available and existing. Available means that the lightpath does not use slots occupied by a different lightpath. Existing means that the lightpath has already been established.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required

Returns:

Name Type Description
lightpath_available_check Tuple[Array, Array, Array, Array]

True if lightpath is available
Source code in xlron/environments/env_funcs.py 
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
@partial(jax.jit, static_argnums=(1,))
def check_lightpath_available_and_existing(state: EnvState, params: EnvParams, action: chex.Array) -> (
        Tuple)[chex.Array, chex.Array, chex.Array, chex.Array]:
    """Check if lightpath is available and existing.
    Available means that the lightpath does not use slots occupied by a different lightpath.
    Existing means that the lightpath has already been established.

    Args:
        state: Environment state
        params: Environment parameters

    Returns:
        lightpath_available_check: True if lightpath is available
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_index, initial_slot_index = process_path_action(state, params, action)
    path = get_paths(params, nodes_sd)[path_index]
    # Get unique lightpath index
    lightpath_index = get_lightpath_index(params, nodes_sd, path_index)
    # Get mask for slots that lightpath will occupy
    # negative numbers used so as not to conflict with lightpath indices
    new_lightpath_mask = vmap_set_path_links(
        jnp.full((params.num_links, 1), -2), path, 0, 1, -1
    )
    path_index_array = state.path_index_array[:, initial_slot_index].reshape(-1, 1)
    masked_path_index_array = jnp.where(
        new_lightpath_mask == -1, path_index_array, -2
    )
    lightpath_mask = jnp.where(
        path_index_array == lightpath_index, -1, -2
    )  # Allow current lightpath
    lightpath_existing_check = jnp.array_equal(lightpath_mask, new_lightpath_mask)  # True if all slots are same
    lightpath_mask = jnp.where(masked_path_index_array == -1, -1, lightpath_mask)  # Allow empty slots
    # True if all slots are same or empty
    lightpath_available_check = jnp.logical_or(
        jnp.array_equal(lightpath_mask, new_lightpath_mask), lightpath_existing_check
    )
    curr_lightpath_capacity = jnp.max(
        jnp.where(new_lightpath_mask == -1, state.link_capacity_array[:, initial_slot_index].reshape(-1, 1), 0)
    )
    return lightpath_available_check, lightpath_existing_check, curr_lightpath_capacity, lightpath_index

check_min_two_nodes_assigned(node_departure_array)

Count negative values on each node (row) in node departure array, sum them, must be 2 or greater. This check is important if e.g. an action contains 2 nodes the same therefore only assigns 1 node. Return False if check passed, True if check failed

Parameters:

Name Type Description Default
node_departure_array Array

Node departure array (N x R) where N is number of nodes and R is number of resources

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
def check_min_two_nodes_assigned(node_departure_array: chex.Array):
    """Count negative values on each node (row) in node departure array, sum them, must be 2 or greater.
    This check is important if e.g. an action contains 2 nodes the same therefore only assigns 1 node.
    Return False if check passed, True if check failed

    Args:
        node_departure_array: Node departure array (N x R) where N is number of nodes and R is number of resources

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.sum(jnp.sum(jnp.where(node_departure_array < 0, 1, 0), axis=1)) <= 1

check_no_spectrum_reuse(link_slot_array)

slot-=1 when used, should be zero when unoccupied, so check if any < -1 in slot array.

Parameters:

Name Type Description Default
link_slot_array

Link slot array (L x S) where L is number of links and S is number of slots

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
def check_no_spectrum_reuse(link_slot_array):
    """slot-=1 when used, should be zero when unoccupied, so check if any < -1 in slot array.

    Args:
        link_slot_array: Link slot array (L x S) where L is number of links and S is number of slots

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.any(link_slot_array < -1)

check_node_capacities(capacity_array)

Sum selected nodes array and check less than node resources.

Parameters:

Name Type Description Default
capacity_array Array

Node capacity array (N x 1) where N is number of nodes

 required

Returns:

Name Type Description
bool bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
def check_node_capacities(capacity_array: chex.Array) -> bool:
    """Sum selected nodes array and check less than node resources.

    Args:
        capacity_array: Node capacity array (N x 1) where N is number of nodes

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.any(capacity_array < 0)

check_snr_sufficient(state, params)

Check if SNR is sufficient for all connections Args: state (EnvState): Environment state params (EnvParams): Environment parameters Returns: jnp.array: 1 if SNR is sufficient for connection else 0

Source code in xlron/environments/env_funcs.py 
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
def check_snr_sufficient(state: RSAGNModelEnvState, params: RSAGNModelEnvParams) -> chex.Array:
    """Check if SNR is sufficient for all connections
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: 1 if SNR is sufficient for connection else 0
    """
    # TODO - this check needs to be faster!
    required_snr_array = get_required_snr_se_kurtosis_array(state.modulation_format_index_array, 2, params)
    # Transform lightpath index array by getting lightpath value, getting path-link array, and summing inverse link SNRs
    lightpath_snr_array = get_lightpath_snr(state, params)
    check_snr_sufficient = jnp.where(lightpath_snr_array >= required_snr_array, 0, 1)
    # jax.debug.print("check_snr_sufficient {}", check_snr_sufficient, ordered=True)
    # jax.debug.print("required_snr_array {}", required_snr_array, ordered=True)
    # jax.debug.print("lightpath_snr_array {}", lightpath_snr_array, ordered=True)
    # jax.debug.print("state.modulation_format_index_array {}", state.modulation_format_index_array, ordered=True)
    # jax.debug.print("state.channel_centre_bw_array {}", state.channel_centre_bw_array, ordered=True)
    # jax.debug.print("state.channel_power_array {}", state.channel_power_array, ordered=True)
    return jnp.any(check_snr_sufficient)

check_topology(action_history, topology_pattern)

Check that each unique virtual node (as indicated by topology pattern) is assigned to a consistent physical node i.e. start and end node of ring is same physical node. Method: For each node index in topology pattern, mask action history with that index, then find max value in masked array. If max value is not the same for all values for that virtual node in action history, then return 1, else 0. Array should be all zeroes at the end, so do an any() check on that. e.g. virtual topology pattern = [2,1,3,1,4,1,2] 3 node ring action history = [0,34,4,0,3,1,0] meaning v node "2" goes p node 0, v node "3" goes p node 4, v node "4" goes p node 3 The numbers in-between relate to the slot action. If any value in the array is 1, a virtual node is assigned to multiple different physical nodes. Need to check from both perspectives: 1. For each virtual node, check that all physical nodes are the same 2. For each physical node, check that all virtual nodes are the same

Parameters:

Name Type Description Default
action_history

Action history

 required
topology_pattern

Topology pattern

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
def check_topology(action_history, topology_pattern):
    """Check that each unique virtual node (as indicated by topology pattern) is assigned to a consistent physical node
    i.e. start and end node of ring is same physical node.
    Method:
    For each node index in topology pattern, mask action history with that index, then find max value in masked array.
    If max value is not the same for all values for that virtual node in action history, then return 1, else 0.
    Array should be all zeroes at the end, so do an any() check on that.
    e.g. virtual topology pattern = [2,1,3,1,4,1,2]  3 node ring
    action history = [0,34,4,0,3,1,0]
    meaning v node "2" goes p node 0, v node "3" goes p node 4, v node "4" goes p node 3
    The numbers in-between relate to the slot action.
    If any value in the array is 1, a virtual node is assigned to multiple different physical nodes.
    Need to check from both perspectives:
    1. For each virtual node, check that all physical nodes are the same
    2. For each physical node, check that all virtual nodes are the same

    Args:
        action_history: Action history
        topology_pattern: Topology pattern

    Returns:
        bool: True if check failed, False if check passed
    """
    def loop_func_virtual(i, val):
        # Get indices of physical node in action history that correspond to virtual node i
        masked_val = jnp.where(i == topology_pattern, val, -1)
        # Get maximum value at those indices (should all be same)
        max_node = jnp.max(masked_val)
        # For relevant indices, if max value then return 0 else 1
        val = jnp.where(masked_val != -1, masked_val != max_node, val)
        return val
    def loop_func_physical(i, val):
        # Get indices of virtual nodes in topology pattern that correspond to physical node i
        masked_val = jnp.where(i == action_history, val, -1)
        # Get maximum value at those indices (should all be same)
        max_node = jnp.max(masked_val)
        # For relevant indices, if max value then return 0 else 1
        val = jnp.where(masked_val != -1, masked_val != max_node, val)
        return val
    topology_pattern = topology_pattern[::2]  # Only look at node indices, not slot actions
    action_history = action_history[::2]
    check_virtual = jax.lax.fori_loop(jnp.min(topology_pattern), jnp.max(topology_pattern)+1, loop_func_virtual, action_history)
    check_physical = jax.lax.fori_loop(jnp.min(action_history), jnp.max(action_history)+1, loop_func_physical, topology_pattern)
    check = jnp.concatenate((check_virtual, check_physical))
    return jnp.any(check)

check_unique_nodes(node_departure_array)

Count negative values on each node (row) in node departure array, must not exceed 1.

Parameters:

Name Type Description Default
node_departure_array Array

Node departure array (N x R) where N is number of nodes and R is number of resources

 required

Returns:

Name Type Description
bool bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
@jax.jit
def check_unique_nodes(node_departure_array: chex.Array) -> bool:
    """Count negative values on each node (row) in node departure array, must not exceed 1.

    Args:
        node_departure_array: Node departure array (N x R) where N is number of nodes and R is number of resources

    Returns:
        bool: True if check failed, False if check passed
    """
    return jnp.any(jnp.sum(jnp.where(node_departure_array < 0, 1, 0), axis=1) > 1)

check_vone_action(state, remaining_actions, total_requested_nodes)

Check if action is valid. Combines checks for: - sufficient node capacities - unique nodes assigned - minimum two nodes assigned - all requested nodes assigned - correct topology pattern - no spectrum reuse

Parameters:

Name Type Description Default
state

current state

 required
remaining_actions

remaining actions

 required
total_requested_nodes

total requested nodes

 required

Returns:

Name Type Description
bool

True if check failed, False if check passed
Source code in xlron/environments/env_funcs.py 
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
def check_vone_action(state, remaining_actions, total_requested_nodes):
    """Check if action is valid.
    Combines checks for:
    - sufficient node capacities
    - unique nodes assigned
    - minimum two nodes assigned
    - all requested nodes assigned
    - correct topology pattern
    - no spectrum reuse

    Args:
        state: current state
        remaining_actions: remaining actions
        total_requested_nodes: total requested nodes

    Returns:
        bool: True if check failed, False if check passed
    """
    checks = jnp.stack((
        check_node_capacities(state.node_capacity_array),
        check_unique_nodes(state.node_departure_array),
        # TODO (VONE) - Remove two nodes check if impairs performance
        #  (check_all_nodes_assigned is sufficient but fails after last action of request instead of earlier)
        check_min_two_nodes_assigned(state.node_departure_array),
        jax.lax.cond(
            jnp.equal(remaining_actions, jnp.array(1)),
            lambda x: check_all_nodes_assigned(*x),
            lambda x: jnp.array(False),
            (state.node_departure_array, total_requested_nodes)
        ),
        jax.lax.cond(
            jnp.equal(remaining_actions, jnp.array(1)),
            lambda x: check_topology(*x),
            lambda x: jnp.array(False),
            (state.action_history, state.request_array[1])
        ),
        check_no_spectrum_reuse(state.link_slot_array),
    ))
    #jax.debug.print("Checks: {}", checks, ordered=True)
    return jnp.any(checks)

convert_node_probs_to_traffic_matrix(node_probs)

Convert list of node probabilities to symmetric traffic matrix.

Parameters:

Name Type Description Default
node_probs list

node probabilities

 required

Returns:

Name Type Description
traffic_matrix Array

traffic matrix
Source code in xlron/environments/env_funcs.py 
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
def convert_node_probs_to_traffic_matrix(node_probs: list) -> chex.Array:
    """Convert list of node probabilities to symmetric traffic matrix.

    Args:
        node_probs: node probabilities

    Returns:
        traffic_matrix: traffic matrix
    """
    matrix = jnp.outer(node_probs, node_probs)
    # Set lead diagonal to zero
    matrix = jnp.where(jnp.eye(matrix.shape[0]) == 1, 0, matrix)
    matrix = normalise_traffic_matrix(matrix)
    return matrix

create_run_name(config)

Create name for run based on config flags

Source code in xlron/environments/env_funcs.py 
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
def create_run_name(config: Union[box.Box, dict]) -> str:
    """Create name for run based on config flags"""
    env_type = config["env_type"]
    topology = config["topology_name"]
    slots = config["link_resources"]
    gnn = "_GNN" if config["USE_GNN"] else ""
    incremental = "_INC" if config["incremental_loading"] else ""
    run_name = f"{env_type}_{topology}_{slots}{gnn}{incremental}".upper()
    if config["EVAL_HEURISTIC"]:
        run_name += f"_{config['path_heuristic']}"
        if env_type.lower() == "vone":
            run_name += f"_{config['node_heuristic']}"
    elif config["EVAL_MODEL"]:
        run_name += f"_EVAL"
    return run_name

decrement_action_counter(state)

Decrement action counter in-place. Used in VONE environments.

Source code in xlron/environments/env_funcs.py 
547
548
549
550
def decrement_action_counter(state):
    """Decrement action counter in-place. Used in VONE environments."""
    state.action_counter.at[-1].add(-1)
    return state

ff_ksp(state, params)

Get the first available slot from all paths Method: Go through action mask and find the first available slot on all paths

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required

Returns:

Type Description
Array

chex.Array: Action
Source code in xlron/heuristics/heuristics.py 
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
@partial(jax.jit, static_argnums=(1,))
def ff_ksp(state: EnvState, params: EnvParams) -> chex.Array:
    """Get the first available slot from all paths
    Method: Go through action mask and find the first available slot on all paths

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters

    Returns:
        chex.Array: Action
    """
    first_slots = first_fit(state, params)
    # Chosen path is the one with the lowest index of first available slot
    path_index = jnp.argmin(first_slots)
    slot_index = first_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

finalise_action_rsa(state, params)

Turn departure times positive.

Parameters:

Name Type Description Default
state EnvState

current state

 required

Returns:

Name Type Description
state

updated state
Source code in xlron/environments/env_funcs.py 
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
@partial(jax.jit, donate_argnums=(0,))
def finalise_action_rsa(state: EnvState, params: Optional[EnvParams]):
    """Turn departure times positive.

    Args:
        state: current state

    Returns:
        state: updated state
    """
    _, requested_datarate = read_rsa_request(state.request_array)
    state = state.replace(
        link_slot_departure_array=make_positive(state.link_slot_departure_array),
        accepted_services=state.accepted_services + 1,
        accepted_bitrate=state.accepted_bitrate + requested_datarate[0],
        total_bitrate=state.total_bitrate + requested_datarate[0]
    )
    return state

finalise_action_rwalr(state, params)

Turn departure times positive.

Parameters:

Name Type Description Default
state EnvState

current state

 required

Returns:

Name Type Description
state

updated state
Source code in xlron/environments/env_funcs.py 
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
@partial(jax.jit, donate_argnums=(0,))
def finalise_action_rwalr(state: EnvState, params: Optional[EnvParams]):
    """Turn departure times positive.

    Args:
        state: current state

    Returns:
        state: updated state
    """
    _, requested_datarate = read_rsa_request(state.request_array)
    state = state.replace(
        link_slot_departure_array=make_positive(state.link_slot_departure_array),
        accepted_services=state.accepted_services + 1,
        accepted_bitrate=state.accepted_bitrate + requested_datarate[0],
        total_bitrate=state.total_bitrate + requested_datarate[0]
    )
    return state

finalise_vone_action(state)

Turn departure times positive.

Parameters:

Name Type Description Default
state

current state

 required

Returns:

Name Type Description
state

updated state
Source code in xlron/environments/env_funcs.py 
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
@partial(jax.jit, donate_argnums=(0,))
def finalise_vone_action(state):
    """Turn departure times positive.

    Args:
        state: current state

    Returns:
        state: updated state
    """
    state = state.replace(
        node_departure_array=make_positive(state.node_departure_array),
        link_slot_departure_array=make_positive(state.link_slot_departure_array),
        accepted_services=state.accepted_services + 1,
        accepted_bitrate=state.accepted_bitrate  # TODO - get sum of bitrates for requested links
    )
    return state

first_fit(state, params)

First-Fit Spectrum Allocation. Returns the first fit slot for each path.

Source code in xlron/heuristics/heuristics.py 
466
467
468
469
470
471
472
473
def first_fit(state: EnvState, params: EnvParams) -> chex.Array:
    """First-Fit Spectrum Allocation. Returns the first fit slot for each path."""
    mask = get_action_mask(state, params)
    # Add a column of ones to the mask to make sure that occupied paths have non-zero index in "first_slots"
    mask = jnp.concatenate((mask, jnp.full((mask.shape[0], 1), 1)), axis=1)
    # Get index of first available slots for each path
    first_slots = jnp.argmax(mask, axis=1)
    return first_slots

format_vone_slot_request(state, action)

Format slot request for VONE action into format (source-node, slot, destination-node).

Parameters:

Name Type Description Default
state EnvState

current state

 required
action Array

action to format

 required

Returns:

Type Description
Array

chex.Array: formatted request
Source code in xlron/environments/env_funcs.py 
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
def format_vone_slot_request(state: EnvState, action: chex.Array) -> chex.Array:
    """Format slot request for VONE action into format (source-node, slot, destination-node).

    Args:
        state: current state
        action: action to format

    Returns:
        chex.Array: formatted request
    """
    remaining_actions = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.action_counter, 2, 1))
    full_request = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.request_array, 0, 1))
    unformatted_request = jax.lax.dynamic_slice_in_dim(full_request, (remaining_actions - 1) * 2, 3)
    node_s = jax.lax.dynamic_slice_in_dim(action, 0, 1)
    requested_slots = jax.lax.dynamic_slice_in_dim(unformatted_request, 1, 1)
    node_d = jax.lax.dynamic_slice_in_dim(action, 2, 1)
    formatted_request = jnp.concatenate((node_s, requested_slots, node_d))
    return formatted_request

generate_arrival_holding_times(key, params)

Generate arrival and holding times based on Poisson distributed events. To understand how sampling from e^-x can be transformed to sample from lambdae^-(x/lambda) see: https://en.wikipedia.org/wiki/Inverse_transform_sampling#Examples Basically, inverse transform sampling is used to sample from a distribution with CDF F(x). The CDF of the exponential distribution (lambdae^-{lambdax}) is F(x) = 1 - e^-{lambdax}. Therefore, the inverse CDF is x = -ln(1-u)/lambda, where u is sample from uniform distribution. Therefore, we need to divide jax.random.exponential() by lambda in order to scale the standard exponential CDF. Experimental histograms of this method compared to random.expovariate() in Python's random library show that the two methods are equivalent. Also see: https://numpy.org/doc/stable/reference/random/generated/numpy.random.exponential.html https://jax.readthedocs.io/en/latest/_autosummary/jax.random.exponential.html

Parameters:

Name Type Description Default
key

PRNG key

 required
params

Environment parameters

 required

Returns:

Name Type Description
arrival_time

Arrival time
holding_time

Holding time
Source code in xlron/environments/env_funcs.py 
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
@partial(jax.jit, static_argnums=(1,))
def generate_arrival_holding_times(key, params):
    """
    Generate arrival and holding times based on Poisson distributed events.
    To understand how sampling from e^-x can be transformed to sample from lambda*e^-(x/lambda) see:
    https://en.wikipedia.org/wiki/Inverse_transform_sampling#Examples
    Basically, inverse transform sampling is used to sample from a distribution with CDF F(x).
    The CDF of the exponential distribution (lambda*e^-{lambda*x}) is F(x) = 1 - e^-{lambda*x}.
    Therefore, the inverse CDF is x = -ln(1-u)/lambda, where u is sample from uniform distribution.
    Therefore, we need to divide jax.random.exponential() by lambda in order to scale the standard exponential CDF.
    Experimental histograms of this method compared to random.expovariate() in Python's random library show that
    the two methods are equivalent.
    Also see: https://numpy.org/doc/stable/reference/random/generated/numpy.random.exponential.html
    https://jax.readthedocs.io/en/latest/_autosummary/jax.random.exponential.html

    Args:
        key: PRNG key
        params: Environment parameters

    Returns:
        arrival_time: Arrival time
        holding_time: Holding time
    """
    key_arrival, key_holding = jax.random.split(key, 2)
    arrival_time = jax.random.exponential(key_arrival, shape=(1,)) \
                   / params.arrival_rate  # Divide because it is rate (lambda)
    if params.truncate_holding_time:
        # For DeepRMSA, need to generate holding times that are less than 2*mean_service_holding_time
        key_holding = jax.random.split(key, 5)
        holding_times = jax.vmap(lambda x: jax.random.exponential(x, shape=(1,)) \
                                * params.mean_service_holding_time)(key_holding)
        holding_times = jnp.where(holding_times < 2*params.mean_service_holding_time, holding_times, 0)
        # Get first non-zero value in holding_times
        non_zero_index = jnp.nonzero(holding_times, size=1)[0][0]
        holding_time = jax.lax.dynamic_slice(jnp.squeeze(holding_times), (non_zero_index,), (1,))
    else:
        holding_time = jax.random.exponential(key_holding, shape=(1,)) \
                       * params.mean_service_holding_time  # Multiply because it is mean (1/lambda)
    return arrival_time, holding_time

generate_vone_request(key, state, params)

Generate a new request for the VONE environment. The request has two rows. The first row shows the node and slot values. The first three elements of the second row show the number of unique nodes, the total number of steps, and the remaining steps. These first three elements comprise the action counter. The remaining elements of the second row show the virtual topology pattern, i.e. the connectivity of the virtual topology.

Source code in xlron/environments/env_funcs.py 
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
@partial(jax.jit, static_argnums=(2,))
def generate_vone_request(key: chex.PRNGKey, state: EnvState, params: EnvParams):
    """Generate a new request for the VONE environment.
    The request has two rows. The first row shows the node and slot values.
    The first three elements of the second row show the number of unique nodes, the total number of steps, and the remaining steps.
    These first three elements comprise the action counter.
    The remaining elements of the second row show the virtual topology pattern, i.e. the connectivity of the virtual topology.
    """
    shape = params.max_edges*2+1  # shape of request array
    key_topology, key_node, key_slot, key_times = jax.random.split(key, 4)
    # Randomly select topology, node resources, slot resources
    pattern = jax.random.choice(key_topology, state.virtual_topology_patterns)
    action_counter = jax.lax.dynamic_slice(pattern, (0,), (3,))
    topology_pattern = jax.lax.dynamic_slice(pattern, (3,), (pattern.shape[0]-3,))
    selected_node_values = jax.random.choice(key_node, state.values_nodes, shape=(shape,))
    selected_bw_values = jax.random.choice(key_slot, params.values_bw.val, shape=(shape,))
    # Create a mask for odd and even indices
    mask = jnp.tile(jnp.array([0, 1]), (shape+1) // 2)[:shape]
    # Vectorized conditional replacement using mask
    first_row = jnp.where(mask, selected_bw_values, selected_node_values)
    # Make sure node request values are consistent for same virtual nodes
    first_row = jax.lax.fori_loop(
        2,  # Lowest node index in virtual topology requests is 2
        shape,  # Highest possible node index in virtual topology requests is shape-1
        lambda i, x: jnp.where(topology_pattern == i, selected_node_values[i], x),
        first_row
    )
    # Mask out unused part of request array
    first_row = jnp.where(topology_pattern == 0, 0, first_row)
    # Set times
    arrival_time, holding_time = generate_arrival_holding_times(key, params)
    state = state.replace(
        holding_time=holding_time,
        current_time=state.current_time + arrival_time,
        action_counter=action_counter,
        request_array=jnp.vstack((first_row, topology_pattern)),
        action_history=init_action_history(params),
        total_requests=state.total_requests + 1
    )
    state = remove_expired_node_requests(state) if not params.incremental_loading else state
    state = remove_expired_services_rsa(state) if not params.incremental_loading else state
    return state

get_action_mask(state, params)

N.B. The mask must already be present in the state!

Source code in xlron/heuristics/heuristics.py 
401
402
403
404
405
406
407
def get_action_mask(state: EnvState, params: EnvParams) -> chex.Array:
    """N.B. The mask must already be present in the state!"""
    mask = state.link_slot_mask if params.__class__.__name__ != "DeepRMSAEnvParams" else (
        mask_slots(state, params, state.request_array).link_slot_mask
    )
    mask = jnp.reshape(mask, (params.k_paths, -1))
    return mask

get_best_modulation_format(state, path, initial_slot_index, launch_power, params)

Get best modulation format for lightpath. "Best" is the highest order that has SNR requirements below available. Try each modulation format, calculate SNR for each, then return the highest order possible. Args: state (EnvState): Environment state path (chex.Array): Path array initial_slot_index (int): Initial slot index params (EnvParams): Environment parameters Returns: jnp.array: Acceptable modulation format indices

Source code in xlron/environments/env_funcs.py 
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
@partial(jax.jit, static_argnums=(3,))
def get_best_modulation_format(state: EnvState, path: chex.Array, initial_slot_index: int, launch_power: chex.Array, params: EnvParams) -> chex.Array:
    """Get best modulation format for lightpath. "Best" is the highest order that has SNR requirements below available.
    Try each modulation format, calculate SNR for each, then return the highest order possible.
    Args:
        state (EnvState): Environment state
        path (chex.Array): Path array
        initial_slot_index (int): Initial slot index
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Acceptable modulation format indices
    """
    _, requested_datarate = read_rsa_request(state.request_array)
    mod_format_count = params.modulations_array.val.shape[0]
    acceptable_mod_format_indices = jnp.full((mod_format_count,), -2)

    def acceptable_modulation_format(i, acceptable_format_indices):
        req_snr = params.modulations_array.val[i][2] + params.snr_margin
        se = params.modulations_array.val[i][1]
        req_slots = required_slots(requested_datarate, se, params.slot_size, params.guardband)
        # TODO - need to check we don't overwrite values in already occupied slots
        # Possible approaches:
        # Check slot occupancy? Probably would need to iterate through for num_slots, but that's an issue
        # What about we allocate and then fix up later, e.g. could it be possible to just add the modulation format on top without
        # check sum of path links prior to assigning?
        #
        new_state = state.replace(
            channel_power_array=vmap_set_path_links(
                state.channel_power_array, path, initial_slot_index, req_slots, launch_power),
            channel_centre_bw_array=vmap_set_path_links(
                state.channel_centre_bw_array, path, initial_slot_index, req_slots, params.slot_size)
        )
        snr_value = get_minimum_snr_of_channels_on_path(new_state, path, initial_slot_index, req_slots, params)
        # jax.debug.print("snr_value {}", snr_value, ordered=True)
        # jax.debug.print("req_snr {}", req_snr, ordered=True)
        acceptable_format_index = jnp.where(snr_value >= req_snr, i, -1).reshape((1,))
        acceptable_format_indices = jax.lax.dynamic_update_slice(acceptable_format_indices, acceptable_format_index, (i,))
        # jax.debug.print("acceptable_format_indices {}", acceptable_format_indices, ordered=True)
        return acceptable_format_indices

    acceptable_mod_format_indices = jax.lax.fori_loop(
        0,
        mod_format_count,
        acceptable_modulation_format,
        acceptable_mod_format_indices
    )
    return acceptable_mod_format_indices

get_best_modulation_format_simple(state, path, initial_slot_index, params)

Get modulation format for lightpath. Assume worst case (least Gaussian) modulation format when calculating SNR. Args: state (EnvState): Environment state path (chex.Array): Path array initial_slot_index (int): Initial slot index params (EnvParams): Environment parameters Returns: jnp.array: Acceptable modulation format indices

Source code in xlron/environments/env_funcs.py 
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
@partial(jax.jit, static_argnums=(3,))
def get_best_modulation_format_simple(
        state: RSAGNModelEnvState, path: chex.Array, initial_slot_index: int, params: RSAGNModelEnvParams
) -> chex.Array:
    """Get modulation format for lightpath.
    Assume worst case (least Gaussian) modulation format when calculating SNR.
    Args:
        state (EnvState): Environment state
        path (chex.Array): Path array
        initial_slot_index (int): Initial slot index
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Acceptable modulation format indices
    """
    link_snr_array = get_snr_link_array(state, params)
    snr_value = get_snr_for_path(path, link_snr_array, params)[initial_slot_index] - params.snr_margin  # Margin
    mod_format_count = params.modulations_array.val.shape[0]
    acceptable_mod_format_indices = jnp.arange(mod_format_count)
    req_snr = params.modulations_array.val[:, 2] + params.snr_margin
    acceptable_mod_format_indices = jnp.where(snr_value >= req_snr,
                                              acceptable_mod_format_indices,
                                              jnp.full((mod_format_count,), -2))
    return acceptable_mod_format_indices

get_centre_frequency(initial_slot_index, num_slots, params)

Get centre frequency for new lightpath

Parameters:

Name Type Description Default
initial_slot_index Array

Centre frequency of first slot

 required
num_slots float

Number of slots

 required
params RSAGNModelEnvParams

Environment parameters

 required

Returns:

Type Description
Array

chex.Array: Centre frequency for new lightpath
Source code in xlron/environments/env_funcs.py 
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
@partial(jax.jit, static_argnums=(2,))
def get_centre_frequency(initial_slot_index: int, num_slots: int, params: RSAGNModelEnvParams) -> chex.Array:
    """Get centre frequency for new lightpath

    Args:
        initial_slot_index (chex.Array): Centre frequency of first slot
        num_slots (float): Number of slots
        params (RSAGNModelEnvParams): Environment parameters

    Returns:
        chex.Array: Centre frequency for new lightpath
    """
    slot_centres = (jnp.arange(params.link_resources) - (params.link_resources + 1) / 2) * params.slot_size
    return slot_centres[initial_slot_index] + (params.slot_size * (num_slots + 1)) / 2

get_edge_disjoint_paths(graph)

Get edge disjoint paths between all nodes in graph.

Parameters:

Name Type Description Default
graph Graph

graph

 required

Returns:

Name Type Description
dict dict

edge disjoint paths (path is list of edges)
Source code in xlron/environments/env_funcs.py 
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
def get_edge_disjoint_paths(graph: nx.Graph) -> dict:
    """Get edge disjoint paths between all nodes in graph.

    Args:
        graph: graph

    Returns:
        dict: edge disjoint paths (path is list of edges)
    """
    result = {n: {} for n in graph}
    for n1, n2 in itertools.combinations(graph, 2):
        # Sort by number of links in path
        # TODO - sort by path length
        result[n1][n2] = sorted(list(nx.edge_disjoint_paths(graph, n1, n2)), key=len)
        result[n2][n1] = sorted(list(nx.edge_disjoint_paths(graph, n2, n1)), key=len)
    return result

get_launch_power(state, path_action, power_action, params)

Get launch power for new lightpath. N.B. launch power is specified in dBm but is converted to linear units when stored in channel_power_array. This func returns linear units (mW). Path action is used to determine the launch power in the case of tabular launch power type. Power action is used to determine the launch power in the case of RL launch power type. During masking, power action is set as state.launch_power_array[0], which is set by the RL agent. Args: state (EnvState): Environment state path_action (chex.Array): Action specifying path index (0 to k_paths-1) power_action (chex.Array): Action specifying launch power in dBm params (EnvParams): Environment parameters Returns: chex.Array: Launch power for new lightpath

Source code in xlron/environments/env_funcs.py 
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
@partial(jax.jit, static_argnums=(1,))
def get_launch_power(state: EnvState, path_action: chex.Array, power_action: chex.Array, params: EnvParams) -> chex.Array:
    """Get launch power for new lightpath. N.B. launch power is specified in dBm but is converted to linear units
    when stored in channel_power_array. This func returns linear units (mW).
    Path action is used to determine the launch power in the case of tabular launch power type.
    Power action is used to determine the launch power in the case of RL launch power type. During masking,
    power action is set as state.launch_power_array[0], which is set by the RL agent.
    Args:
        state (EnvState): Environment state
        path_action (chex.Array): Action specifying path index (0 to k_paths-1)
        power_action (chex.Array): Action specifying launch power in dBm
        params (EnvParams): Environment parameters
    Returns:
        chex.Array: Launch power for new lightpath
    """
    k_path_index, _ = process_path_action(state, params, path_action)
    if params.launch_power_type == 1:  # Fixed
        return state.launch_power_array[0]
    elif params.launch_power_type == 2:  # Tabular (one row per path)
        nodes_sd, requested_datarate = read_rsa_request(state.request_array)
        source, dest = nodes_sd
        i = get_path_indices(source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph).astype(
            jnp.int32)
        return state.launch_power_array[i+k_path_index]
    elif params.launch_power_type == 3:  # RL
        return power_action
    elif params.launch_power_type == 4:  # Scaled
        nodes_sd, requested_datarate = read_rsa_request(state.request_array)
        source, dest = nodes_sd
        i = get_path_indices(
            source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph
        ).astype(jnp.int32)
        # Get path length
        link_length_array = jnp.sum(params.link_length_array.val, axis=1)
        path_length = jnp.sum(link_length_array[i+k_path_index])
        maximum_path_length = jnp.max(jnp.dot(params.path_link_array.val, params.link_length_array.val))
        return state.launch_power_array[0] * (path_length / maximum_path_length)
    else:
        raise ValueError("Invalid launch power type. Check params.launch_power_type")

get_lightpath_snr(state, params)

Get SNR for each link on path. N.B. that in most cases it is more efficient to calculate the SNR for every possible path, rather than a slot-by-slot basis. But in some cases slot-by-slot is better i.e. when kN(N-1)/2 > LS Args: state (RSAGNModelEnvState): Environment state params (RSAGNModelEnvParams): Environment parameters

Returns:

Type Description
Array

chex.array: SNR for each link on path
Source code in xlron/environments/env_funcs.py 
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
def get_lightpath_snr(state: RSAGNModelEnvParams, params: RSAGNModelEnvParams) -> chex.Array:
    """Get SNR for each link on path.
    N.B. that in most cases it is more efficient to calculate the SNR for every possible path, rather than a slot-by-slot basis.
    But in some cases slot-by-slot is better i.e. when k*N(N-1)/2 > L*S
    Args:
        state (RSAGNModelEnvState): Environment state
        params (RSAGNModelEnvParams): Environment parameters

    Returns:
        chex.array: SNR for each link on path
    """
    # Get the SNR for the channel that the path occupies
    path_snr_array = jax.vmap(get_snr_for_path, in_axes=(0, None, None))(params.path_link_array.val, state.link_snr_array, params)
    # Where value in path_index_array matches index of path_snr_array, substitute in SNR value
    slot_indices = jnp.arange(params.link_resources)
    lightpath_snr_array = jax.vmap(jax.vmap(lambda x, si: path_snr_array[x][si], in_axes=(0, 0)), in_axes=(0, None))(state.path_index_array, slot_indices)
    return lightpath_snr_array

Get link weights based on occupancy for use in congestion-aware routing heuristics.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required

Returns:

Type Description

chex.Array: Link weights
Source code in xlron/heuristics/heuristics.py 
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
def get_link_weights(state: EnvState, params: EnvParams):
    """Get link weights based on occupancy for use in congestion-aware routing heuristics.

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters

    Returns:
        chex.Array: Link weights
    """
    if params.__class__.__name__ != "RWALightpathReuseEnvParams":
        link_occupancy = jnp.count_nonzero(state.link_slot_array, axis=1)
    else:
        initial_path_capacity = init_path_capacity_array(
            params.link_length_array.val, params.path_link_array.val, scale_factor=1.0
        )
        initial_path_capacity = jnp.squeeze(jax.vmap(lambda x: initial_path_capacity[x])(state.path_index_array))
        utilisation = jnp.where(initial_path_capacity - state.link_capacity_array < 0, 0,
                                initial_path_capacity - state.link_capacity_array) / initial_path_capacity
        link_occupancy = jnp.sum(utilisation, axis=1)
    link_weights = jnp.multiply(params.link_length_array.val.T, (1 / (1 - link_occupancy / (params.link_resources + 1))))[0]
    return link_weights

get_minimum_snr_of_channels_on_path(state, path, slot_index, req_slots, params)

Get the minimum value of the SNR on newly assigned channels. N.B. this requires the link_snr_array to have already been calculated and present in state.

Source code in xlron/environments/env_funcs.py 
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
@partial(jax.jit, static_argnums=(2,))
def get_minimum_snr_of_channels_on_path(
        state: RSAGNModelEnvState, path: chex.Array, slot_index: chex.Array, req_slots: int, params: RSAGNModelEnvParams
) -> chex.Array:
    """Get the minimum value of the SNR on newly assigned channels.
    N.B. this requires the link_snr_array to have already been calculated and present in state."""
    snr_value_all_channels = get_snr_for_path(path, state.link_snr_array, params)
    min_snr_value_sub_channels = jnp.min(
        jnp.concatenate([
            snr_value_all_channels[slot_index].reshape((1,)),
            snr_value_all_channels[slot_index + req_slots - 1].reshape((1,))
        ], axis=0)
    )
    return min_snr_value_sub_channels

get_num_spectral_features(n_nodes)

Heuristic for number of spectral features based on graph size.

Parameters:

Name Type Description Default
n_nodes int

Number of nodes in the graph

 required

Returns:

Type Description
int

Number of spectral features to use, clamped between 3 and 15.
int

Follows log2(n_nodes) scaling as reasonable default.
Source code in xlron/environments/env_funcs.py 
35
36
37
38
39
40
41
42
43
44
45
46
@partial(jax.jit, static_argnums=(0,))
def get_num_spectral_features(n_nodes: int) -> int:
    """Heuristic for number of spectral features based on graph size.

    Args:
        n_nodes: Number of nodes in the graph

    Returns:
        Number of spectral features to use, clamped between 3 and 15.
        Follows log2(n_nodes) scaling as reasonable default.
    """
    return jnp.minimum(jnp.maximum(3, jnp.floor(jnp.log2(n_nodes))), 15).astype(int)

get_path_from_path_index_array(path_index_array, path_link_array)

Get path from path index array. Args: path_index_array (chex.Array): Path index array path_link_array (chex.Array): Path link array

Returns:

Type Description
Array

jnp.array: path index values replaced with binary path-link arrays
Source code in xlron/environments/env_funcs.py 
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
@partial(jax.jit, static_argnums=(1,))
def get_path_from_path_index_array(path_index_array: chex.Array, path_link_array: chex.Array) -> chex.Array:
    """Get path from path index array.
    Args:
        path_index_array (chex.Array): Path index array
        path_link_array (chex.Array): Path link array

    Returns:
        jnp.array: path index values replaced with binary path-link arrays
    """
    def get_index_from_link(link):
        return jax.vmap(lambda x: path_link_array[x], in_axes=(0,))(link)

    return jax.vmap(get_index_from_link, in_axes=(0,))(path_index_array)

get_path_index_array(params, nodes)

Indices of paths between source and destination from path array

Source code in xlron/environments/env_funcs.py 
742
743
744
745
746
747
748
749
@partial(jax.jit, static_argnums=(0,))
def get_path_index_array(params, nodes):
    """Indices of paths between source and destination from path array"""
    # get source and destination nodes in order (for accurate indexing of path-link array)
    source, dest = nodes
    i = get_path_indices(source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph).astype(jnp.int32)
    index_array = jax.lax.dynamic_slice(jnp.arange(0, params.path_link_array.shape[0]), (i,), (params.k_paths,))
    return index_array

get_path_indices(s, d, k, N, directed=False)

Get path indices for a given source, destination and number of paths. If source > destination and the graph is directed (two fibres per link, one in each direction) then an offset is added to the index to get the path in the other direction (the offset is the total number source-dest pairs).

Parameters:

Name Type Description Default
s int

Source node index

 required
d int

Destination node index

 required
k int

Number of paths

 required
N int

Number of nodes

 required
directed bool

Whether graph is directed. Defaults to False.

 False

Returns:

Type Description
Array

jnp.array: Start index on path-link array for candidate paths
Source code in xlron/environments/env_funcs.py 
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
@partial(jax.jit, static_argnums=(2, 3, 4))
def get_path_indices(s: int, d: int, k: int, N: int, directed: bool = False) -> chex.Array:
    """Get path indices for a given source, destination and number of paths.
    If source > destination and the graph is directed (two fibres per link, one in each direction) then an offset is
    added to the index to get the path in the other direction (the offset is the total number source-dest pairs).

    Args:
        s (int): Source node index
        d (int): Destination node index
        k (int): Number of paths
        N (int): Number of nodes
        directed (bool, optional): Whether graph is directed. Defaults to False.

    Returns:
        jnp.array: Start index on path-link array for candidate paths
    """
    node_indices = jnp.arange(N, dtype=jnp.int32)
    indices_to_s = jnp.where(node_indices < s, node_indices, 0)
    indices_to_d = jnp.where(node_indices < d, node_indices, 0)
    # If two fibres per link, add offset to index to get fibre in other direction if source > destination
    directed_offset = directed * (s > d) * N * (N - 1) * k / 2
    # The following equation is based on the combinations formula
    forward = ((N * s + d - jnp.sum(indices_to_s) - 2 * s - 1) * k)
    backward = ((N * d + s - jnp.sum(indices_to_d) - 2 * d - 1) * k)
    return forward * (s < d) + backward * (s > d) + directed_offset

get_path_slots(link_slot_array, params, nodes_sd, i, agg_func='max')

Get slots on each constitutent link of path from link_slot_array (L x S), then aggregate to get (S x 1) representation of slots on path.

Parameters:

Name Type Description Default
link_slot_array Array

link-slot array

 required
params EnvParams

environment parameters

 required
nodes_sd Array

source-destination nodes

 required
i int

path index

 required
agg_func str

aggregation function (max or sum). If max, result will be available slots on path. If sum, result will contain information on edge features.

 'max'

Returns:

Name Type Description
slots Array

slots on path
Source code in xlron/environments/env_funcs.py 
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
@partial(jax.jit, static_argnums=(1, 4))
def get_path_slots(link_slot_array: chex.Array, params: EnvParams, nodes_sd: chex.Array, i: int, agg_func: str = "max") -> chex.Array:
    """Get slots on each constitutent link of path from link_slot_array (L x S),
    then aggregate to get (S x 1) representation of slots on path.

    Args:
        link_slot_array: link-slot array
        params: environment parameters
        nodes_sd: source-destination nodes
        i: path index
        agg_func: aggregation function (max or sum).
            If max, result will be available slots on path.
            If sum, result will contain information on edge features.

    Returns:
        slots: slots on path
    """
    path = get_paths(params, nodes_sd)[i]
    path = path.reshape((params.num_links, 1))
    # Get links and collapse to single dimension
    num_slots = params.link_resources if agg_func == "max" else math.ceil(params.link_resources/params.aggregate_slots)
    slots = jnp.where(path, link_slot_array, jnp.zeros(num_slots))
    # Make any -1s positive then get max for each slot across links
    if agg_func == "max":
        # Use this for getting slots from link_slot_array
        slots = jnp.max(jnp.absolute(slots), axis=0)
    elif agg_func == "sum":
        # TODO - consider using an RNN (or S5) to aggregate edge features
        # Use this (or alternative) for aggregating edge features from GNN
        slots = jnp.sum(slots, axis=0)
    else:
        raise ValueError("agg_func must be 'max' or 'sum'")
    return slots

get_paths(params, nodes)

Get k paths between source and destination

Source code in xlron/environments/env_funcs.py 
752
753
754
755
756
@partial(jax.jit, static_argnums=(0,))
def get_paths(params, nodes):
    """Get k paths between source and destination"""
    index_array = get_path_index_array(params, nodes)
    return jnp.take(params.path_link_array.val, index_array, axis=0)

get_paths_obs_gn_model(state, params)

Get observation space for launch power optimization (with numerical stability).

Source code in xlron/environments/env_funcs.py 
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
@partial(jax.jit, static_argnums=(1,))
def get_paths_obs_gn_model(state: RSAGNModelEnvState, params: RSAGNModelEnvParams) -> chex.Array:
    # TODO - make this just show the stats from just one path at a time
    """Get observation space for launch power optimization (with numerical stability)."""
    request_array = state.request_array.reshape((-1,))
    path_stats = calculate_path_stats(state, params, request_array)
    # Remove first 3 items of path stats for each path
    path_stats = path_stats[:, 3:]
    link_length_array = jnp.sum(params.link_length_array.val, axis=1)
    lightpath_snr_array = get_lightpath_snr(state, params)
    nodes_sd, requested_datarate = read_rsa_request(request_array)
    source, dest = nodes_sd

    def calculate_gn_path_stats(k_path_index, init_val):
        # Get path index
        path_index = get_path_indices(source, dest, params.k_paths, params.num_nodes,
                                      directed=params.directed_graph).astype(
            jnp.int32) + k_path_index
        path = params.path_link_array[path_index]
        path_length = jnp.dot(path, link_length_array)
        max_path_length = jnp.max(jnp.dot(params.path_link_array.val, link_length_array))
        path_length_norm = path_length / max_path_length
        path_length_hops_norm = jnp.sum(path) / jnp.max(jnp.sum(params.path_link_array.val, axis=1))
        # Connections on path
        num_connections = jnp.where(path == 1, jnp.where(state.channel_power_array > 0, 1, 0).sum(axis=1), 0).sum()
        num_connections_norm = num_connections / params.link_resources
        # Mean power of connections on path
        # make path with row length equal to link_resource (+1 to avoid zero division)
        mean_power_norm = (jnp.where(path == 1, state.channel_power_array.sum(axis=1), 0).sum() /
                           (jnp.where(num_connections > 0., num_connections, 1.) * params.max_power))
        # Mean SNR of connections on the path links
        max_snr = 50.  # Nominal value for max GSNR in dB
        mean_snr_norm = (jnp.where(path == 1, lightpath_snr_array.sum(axis=1), 0).sum() /
                         (jnp.where(num_connections > 0., num_connections, 1.) * max_snr))
        return jax.lax.dynamic_update_slice(
            init_val,
            jnp.array([[
                path_length,
                path_length_hops_norm,
                num_connections_norm,
                mean_power_norm,
                mean_snr_norm
            ]]),
            (k_path_index, 0),
        )

    gn_path_stats = jnp.zeros((params.k_paths, 5))
    gn_path_stats = jax.lax.fori_loop(
        0, params.k_paths, calculate_gn_path_stats, gn_path_stats
    )
    all_stats = jnp.concatenate([path_stats, gn_path_stats], axis=1)
    return jnp.concatenate(
        (
            jnp.array([source]),
            requested_datarate / 100.,
            jnp.array([dest]),
            jnp.reshape(state.holding_time, (-1,)),
            jnp.reshape(all_stats, (-1,)),
        ),
        axis=0,
    )

get_paths_se(params, nodes)

Get max. spectral efficiency of modulation format on k paths between source and destination

Source code in xlron/environments/env_funcs.py 
759
760
761
762
763
764
@partial(jax.jit, static_argnums=(0,))
def get_paths_se(params, nodes):
    """Get max. spectral efficiency of modulation format on k paths between source and destination"""
    # get source and destination nodes in order (for accurate indexing of path-link array)
    index_array = get_path_index_array(params, nodes)
    return jnp.take(params.path_se_array.val, index_array, axis=0)

get_required_snr_se_kurtosis_array(modulation_format_index_array, col_index, params)

Convert modulation format index to required SNR or spectral efficiency. Modulation format index array contains the index of the modulation format used by the channel. The modulation index references a row in the modulations array, which contains SNR and SE values.

Parameters:

Name Type Description Default
modulation_format_index_array Array

Modulation format index array

 required
col_index int

Column index for required SNR or spectral efficiency

 required
params RSAGNModelEnvParams

Environment parameters

 required

Returns:

Type Description
Array

jnp.array: Required SNR for each channel (min. SNR for empty channel (mod. index 0))
Source code in xlron/environments/env_funcs.py 
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
@partial(jax.jit, static_argnums=(1, 2,))
def get_required_snr_se_kurtosis_array(modulation_format_index_array: chex.Array, col_index: int, params: RSAGNModelEnvParams) -> chex.Array:
    """Convert modulation format index to required SNR or spectral efficiency.
    Modulation format index array contains the index of the modulation format used by the channel.
    The modulation index references a row in the modulations array, which contains SNR and SE values.

    Args:
        modulation_format_index_array (chex.Array): Modulation format index array
        col_index (int): Column index for required SNR or spectral efficiency
        params (RSAGNModelEnvParams): Environment parameters

    Returns:
        jnp.array: Required SNR for each channel (min. SNR for empty channel (mod. index 0))
    """
    return jax.vmap(get_required_snr_se_kurtosis_on_link, in_axes=(0, None, None))(modulation_format_index_array, col_index, params)

Get SNR per link Args: state (EnvState): Environment state params (EnvParams): Environment parameters Returns: jnp.array: SNR per link

Source code in xlron/environments/env_funcs.py 
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
@partial(jax.jit, static_argnums=(1,))
def get_snr_link_array(state: EnvState, params: EnvParams) -> chex.Array:
    """Get SNR per link
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: SNR per link
    """

    def get_link_snr(link_index, state, params):
        # Get channel power, channel centre, bandwidth, and noise figure
        link_lengths = params.link_length_array[link_index, :]
        num_spans = jnp.ceil(jnp.sum(link_lengths)*1e3 / params.max_span_length).astype(jnp.int32)
        if params.mod_format_correction:
            mod_format_link = state.modulation_format_index_array[link_index, :]
            kurtosis_link = get_required_snr_se_kurtosis_on_link(mod_format_link, 4, params)
            se_link = get_required_snr_se_kurtosis_on_link(mod_format_link, 1, params)
        else:
            kurtosis_link = jnp.zeros(params.link_resources)
            se_link = jnp.ones(params.link_resources)
        bw_link = state.channel_centre_bw_array[link_index, :]
        ch_power_link = state.channel_power_array[link_index, :]
        required_slots_link = get_required_slots_on_link(bw_link, se_link, params)
        ch_centres_link = get_centre_freq_on_link(jnp.arange(params.link_resources), required_slots_link, params)

        # Calculate SNR
        P = dict(
            num_channels=params.link_resources,
            num_spans=num_spans,
            max_spans=params.max_spans,
            ref_lambda=params.ref_lambda,
            length=link_lengths,
            attenuation_i=jnp.array([params.attenuation]),
            attenuation_bar_i=jnp.array([params.attenuation_bar]),
            nonlinear_coeff=jnp.array([params.nonlinear_coeff]),
            raman_gain_slope_i=jnp.array([params.raman_gain_slope]),
            dispersion_coeff=jnp.array([params.dispersion_coeff]),
            dispersion_slope=jnp.array([params.dispersion_slope]),
            coherent=params.coherent,
            num_roadms=params.num_roadms,
            roadm_loss=params.roadm_loss,
            noise_figure=params.noise_figure,  # TODO (GN MODEL) - support separate noise figures for C and L bands (4 and 6 dB)
            mod_format_correction=params.mod_format_correction,
            ch_power_w_i=ch_power_link.reshape((params.link_resources, 1)),
            ch_centre_i=ch_centres_link.reshape((params.link_resources, 1))*1e9,
            ch_bandwidth_i=bw_link.reshape((params.link_resources, 1))*1e9,
            excess_kurtosis_i=kurtosis_link.reshape((params.link_resources, 1)),
        )
        snr = isrs_gn_model.get_snr(**P)[0]

        return snr

    link_snr_array = jax.vmap(get_link_snr, in_axes=(0, None, None))(jnp.arange(params.num_links), state, params)
    link_snr_array = jnp.nan_to_num(link_snr_array, nan=1e-5)
    return link_snr_array

get_spectral_features(laplacian, num_features)

Compute spectral node features from symmetric normalized graph Laplacian.

Parameters:

Name Type Description Default
adj

Adjacency matrix of the graph

 required
num_features int

Number of eigenvector features to extract

 required

Returns:

Type Description
ndarray

Array of shape (n_nodes, num_features) containing eigenvectors corresponding
ndarray

to the smallest non-zero eigenvalues of the graph Laplacian.
Notes
  • Skips trivial eigenvectors (those with near-zero eigenvalues)
  • Eigenvectors are ordered by ascending eigenvalue magnitude
  • Runtime is O(n^3) - use only for small/medium graphs
  • Eigenvector signs are arbitrary (may vary between runs)
Source code in xlron/environments/env_funcs.py 
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
def get_spectral_features(laplacian: jnp.array, num_features: int) -> jnp.ndarray:
    """Compute spectral node features from symmetric normalized graph Laplacian.

    Args:
        adj: Adjacency matrix of the graph
        num_features: Number of eigenvector features to extract

    Returns:
        Array of shape (n_nodes, num_features) containing eigenvectors corresponding
        to the smallest non-zero eigenvalues of the graph Laplacian.

    Notes:
        - Skips trivial eigenvectors (those with near-zero eigenvalues)
        - Eigenvectors are ordered by ascending eigenvalue magnitude
        - Runtime is O(n^3) - use only for small/medium graphs
        - Eigenvector signs are arbitrary (may vary between runs)
    """
    n_nodes = laplacian.shape[0]
    eigenvalues, eigenvectors = jnp.linalg.eigh(laplacian)
    # Skip trivial eigenvectors (zero eigenvalues for connected components)
    num_zero = jnp.sum(jnp.abs(eigenvalues) < 1e-5)
    valid_features = jnp.minimum(num_features, n_nodes - num_zero)
    return eigenvectors[:, :num_features]#num_zero:num_zero + valid_features]

implement_action_rmsa_gn_model(state, action, params)

Implement action for RSA GN model. Update following arrays: - link_slot_array - link_slot_departure_array - link_snr_array - modulation_format_index_array - channel_power_array - active_path_array Args: state (EnvState): Environment state action (chex.Array): Action tuple (first is path action, second is launch_power) params (EnvParams): Environment parameters Returns: EnvState: Updated environment state

Source code in xlron/environments/env_funcs.py 
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
@partial(jax.jit, static_argnums=(2,))
def implement_action_rmsa_gn_model(
        state: RSAGNModelEnvState, action: chex.Array, params: RSAGNModelEnvParams
) -> EnvState:
    """Implement action for RSA GN model. Update following arrays:
    - link_slot_array
    - link_slot_departure_array
    - link_snr_array
    - modulation_format_index_array
    - channel_power_array
    - active_path_array
    Args:
        state (EnvState): Environment state
        action (chex.Array): Action tuple (first is path action, second is launch_power)
        params (EnvParams): Environment parameters
    Returns:
        EnvState: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_action, power_action = action
    path_action = path_action.astype(jnp.int32)
    k_path_index, initial_slot_index = process_path_action(state, params, path_action)
    lightpath_index = get_lightpath_index(params, nodes_sd, k_path_index)
    path = get_paths(params, nodes_sd)[k_path_index]
    launch_power = get_launch_power(state, path_action, power_action, params)
    # TODO(GN MODEL) - get mod. format based on maximum reach
    mod_format_index = jax.lax.dynamic_slice(
        state.mod_format_mask, (path_action,), (1,)
    ).astype(jnp.int32)[0]
    se = params.modulations_array.val[mod_format_index][1]
    num_slots = required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband)
    # Update link_slot_array and link_slot_departure_array, then other arrays
    state = implement_path_action(state, path, initial_slot_index, num_slots)
    state = state.replace(
        path_index_array=vmap_set_path_links(state.path_index_array, path, initial_slot_index, num_slots, lightpath_index),
        channel_power_array=vmap_set_path_links(state.channel_power_array, path, initial_slot_index, num_slots, launch_power),
        modulation_format_index_array=vmap_set_path_links(state.modulation_format_index_array, path, initial_slot_index, num_slots, mod_format_index),
        channel_centre_bw_array=vmap_set_path_links(state.channel_centre_bw_array, path, initial_slot_index, num_slots, params.slot_size),
    )
    # Update link_snr_array
    state = state.replace(link_snr_array=get_snr_link_array(state, params))
    # jax.debug.print("launch_power {}", launch_power, ordered=True)
    # jax.debug.print("mod_format_index {}", mod_format_index, ordered=True)
    # jax.debug.print("initial_slot_index {}", initial_slot_index, ordered=True)
    # jax.debug.print("state.mod_format_mask {}", state.mod_format_mask, ordered=True)
    # jax.debug.print("path_snr {}", get_snr_for_path(path, state.link_snr_array, params), ordered=True)
    # jax.debug.print("required_snr {}", params.modulations_array.val[mod_format_index][2] + params.snr_margin, ordered=True)
    return state

implement_action_rsa(state, action, params)

Implement action to assign slots on links.

Parameters:

Name Type Description Default
state EnvState

current state

 required
action Array

action to implement

 required
params EnvParams

environment parameters

 required

Returns:

Name Type Description
state EnvState

updated state
Source code in xlron/environments/env_funcs.py 
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
@partial(jax.jit, static_argnums=(2,))
def implement_action_rsa(
        state: EnvState,
        action: chex.Array,
        params: EnvParams,
) -> EnvState:
    """Implement action to assign slots on links.

    Args:
        state: current state
        action: action to implement
        params: environment parameters

    Returns:
        state: updated state
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_index, initial_slot_index = process_path_action(state, params, action)
    path = get_paths(params, nodes_sd)[path_index]
    if params.__class__.__name__ == "RWALightpathReuseEnvParams":
        state = state.replace(
            link_capacity_array=vmap_update_path_links(
                state.link_capacity_array, path, initial_slot_index, 1, requested_datarate
            )
        )
        # TODO (Dynamic-RWALR) - to support diverse requested_datarates for RWA-LR, need to update masking
        # TODO (Dynamic-RWALR) - In order to enable dynamic RWA with lightpath reuse (as opposed to just incremental loading),
        #  need to keep track of active requests OR just randomly remove connections
        #  (could do this by using the link_slot_departure array in a novel way... i.e. don't fill it with departure time but current bw)
        capacity_mask = jnp.where(state.link_capacity_array <= 0., -1., 0.)
        over_capacity_mask = jnp.where(state.link_capacity_array < 0., -1., 0.)
        total_mask = capacity_mask + over_capacity_mask
        state = state.replace(
            link_slot_array=total_mask,
            link_slot_departure_array=vmap_update_path_links(state.link_slot_departure_array, path,
                                                                       initial_slot_index, 1,
                                                                       state.current_time + state.holding_time)
        )
    else:
        se = get_paths_se(params, nodes_sd)[path_index] if params.consider_modulation_format else 1
        num_slots = required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband)
        state = implement_path_action(state, path, initial_slot_index, num_slots)
    return state

implement_action_rsa_gn_model(state, action, params)

Implement action for RSA GN model. Update following arrays: - link_slot_array - link_slot_departure_array - link_snr_array - modulation_format_index_array - channel_power_array - active_path_array Args: state (EnvState): Environment state action (chex.Array): Action tuple (first is path action, second is launch_power) params (EnvParams): Environment parameters Returns: EnvState: Updated environment state

Source code in xlron/environments/env_funcs.py 
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
@partial(jax.jit, static_argnums=(2,))
def implement_action_rsa_gn_model(
        state: RSAGNModelEnvState, action: chex.Array, params: RSAGNModelEnvParams
) -> EnvState:
    """Implement action for RSA GN model. Update following arrays:
    - link_slot_array
    - link_slot_departure_array
    - link_snr_array
    - modulation_format_index_array
    - channel_power_array
    - active_path_array
    Args:
        state (EnvState): Environment state
        action (chex.Array): Action tuple (first is path action, second is launch_power)
        params (EnvParams): Environment parameters
    Returns:
        EnvState: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_action, power_action = action
    path_action = path_action.astype(jnp.int32)
    k_path_index, initial_slot_index = process_path_action(state, params, path_action)
    lightpath_index = get_lightpath_index(params, nodes_sd, k_path_index)
    path = get_paths(params, nodes_sd)[k_path_index]
    launch_power = get_launch_power(state, path_action, power_action, params)
    num_slots = required_slots(requested_datarate, 1, params.slot_size, guardband=params.guardband)
    # Update link_slot_array and link_slot_departure_array, then other arrays
    state = implement_path_action(state, path, initial_slot_index, num_slots)
    state = state.replace(
        path_index_array=vmap_set_path_links(state.path_index_array, path, initial_slot_index, num_slots, lightpath_index),
        channel_power_array=vmap_set_path_links(state.channel_power_array, path, initial_slot_index, num_slots, launch_power),
        channel_centre_bw_array=vmap_set_path_links(state.channel_centre_bw_array, path, initial_slot_index, num_slots, params.slot_size),
        active_lightpaths_array=update_active_lightpaths_array(state, lightpath_index),
        active_lightpaths_array_departure=update_active_lightpaths_array_departure(state, -state.current_time-state.holding_time),
    )
    # Update link_snr_array
    state = state.replace(link_snr_array=get_snr_link_array(state, params))
    return state

implement_action_rwalr(state, action, params)

For use in RWALightpathReuseEnv. Update link_slot_array and link_slot_departure_array to reflect new lightpath assignment. Update link_capacity_array with new capacity if lightpath is available. Undo link_capacity_update if over capacity.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
action Array

Action array

 required
params EnvParams

Environment parameters

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
@partial(jax.jit, static_argnums=(2,))
def implement_action_rwalr(state: EnvState, action: chex.Array, params: EnvParams) -> EnvState:
    """For use in RWALightpathReuseEnv.
    Update link_slot_array and link_slot_departure_array to reflect new lightpath assignment.
    Update link_capacity_array with new capacity if lightpath is available.
    Undo link_capacity_update if over capacity.

    Args:
        state: Environment state
        action: Action array
        params: Environment parameters

    Returns:
        state: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(state.request_array)
    path_index, initial_slot_index = process_path_action(state, params, action)
    path = get_paths(params, nodes_sd)[path_index]
    lightpath_available_check, lightpath_existing_check, curr_lightpath_capacity, lightpath_index = (
        check_lightpath_available_and_existing(state, params, action)
    )
    # Get path capacity - request
    lightpath_capacity = jax.lax.cond(
        lightpath_existing_check,
        lambda x: curr_lightpath_capacity - requested_datarate,  # Subtract requested_datarate from current lightpath
        lambda x: jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.path_capacity_array, x, 1)) - requested_datarate,  # Get initial capacity of lightpath - request
        lightpath_index
    )
    # Update link_capacity_array with new capacity if lightpath is available
    state = jax.lax.cond(
        lightpath_available_check,
        lambda x: x.replace(
            link_capacity_array=vmap_set_path_links(
                state.link_capacity_array, path, initial_slot_index, 1, lightpath_capacity
            ),
            path_index_array=vmap_set_path_links(
                state.path_index_array, path, initial_slot_index, 1, lightpath_index
            ),
        ),
        lambda x: x,
        state
    )
    capacity_mask = jnp.where(state.link_capacity_array <= 0., -1., 0.)
    over_capacity_mask = jnp.where(state.link_capacity_array < 0., -1., 0.)
    # Undo link_capacity_update if over capacity
    # N.B. this will fail if requested capacity is greater than total original capacity of lightpath
    lightpath_capacity_before_action = jax.lax.cond(
        lightpath_existing_check,
        lambda x: curr_lightpath_capacity,  # Subtract requested_datarate from current lightpath
        lambda x: 1e6,  # Empty slots have high capacity (1e6)
        # Get initial capacity of lightpath - request
        None,
    )
    state = state.replace(
        link_capacity_array=jnp.where(over_capacity_mask == -1, lightpath_capacity_before_action, state.link_capacity_array)
    )
    # Total mask will be 0 if space still available, -1 if capacity is zero or -2 if over capacity
    total_mask = capacity_mask + over_capacity_mask
    # Update link_slot_array and link_slot_departure_array
    state = state.replace(
        link_slot_array=total_mask,
        link_slot_departure_array=vmap_update_path_links(state.link_slot_departure_array, path,
                                                                 initial_slot_index, 1,
                                                                 state.current_time + state.holding_time)
    )
    return state

implement_node_action(state, s_node, d_node, s_request, d_request, n=2)

Update node capacity, node resource and node departure arrays

Parameters:

Name Type Description Default
state State

current state

 required
s_node int

source node

 required
d_node int

destination node

 required
s_request int

source node request

 required
d_request int

destination node request

 required
n int

number of nodes to implement. Defaults to 2.

 2

Returns:

Name Type Description
State EnvState

updated state
Source code in xlron/environments/env_funcs.py 
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
def implement_node_action(state: EnvState, s_node: chex.Array, d_node: chex.Array, s_request: chex.Array, d_request: chex.Array, n=2) -> EnvState:
    """Update node capacity, node resource and node departure arrays

    Args:
        state (State): current state
        s_node (int): source node
        d_node (int): destination node
        s_request (int): source node request
        d_request (int): destination node request
        n (int, optional): number of nodes to implement. Defaults to 2.

    Returns:
        State: updated state
    """
    node_indices = jnp.arange(state.node_capacity_array.shape[0])

    curr_selected_nodes = jnp.zeros(state.node_capacity_array.shape[0])
    # d_request -ve so that selected node is +ve (so that argmin works correctly for node resource array update)
    # curr_selected_nodes is N x 1 array, with requested node resources at index of selected node
    curr_selected_nodes = update_node_array(node_indices, curr_selected_nodes, d_node, -d_request)
    curr_selected_nodes = jax.lax.cond(n == 2, lambda x: update_node_array(*x), lambda x: x[1], (node_indices, curr_selected_nodes, s_node, -s_request))

    node_capacity_array = state.node_capacity_array - curr_selected_nodes

    node_resource_array = vmap_update_node_resources(state.node_resource_array, curr_selected_nodes)

    node_departure_array = vmap_update_node_departure(state.node_departure_array, curr_selected_nodes, -state.current_time-state.holding_time)

    state = state.replace(
        node_capacity_array=node_capacity_array,
        node_resource_array=node_resource_array,
        node_departure_array=node_departure_array
    )

    return state

implement_path_action(state, path, initial_slot_index, num_slots)

Update link-slot and link-slot departure arrays. Times are set to negative until turned positive by finalisation (after checks).

Parameters:

Name Type Description Default
state State

current state

 required
path int

path to implement

 required
initial_slot_index int

initial slot index

 required
num_slots int

number of slots to implement

 required
Source code in xlron/environments/env_funcs.py 
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
def implement_path_action(state: EnvState, path: chex.Array, initial_slot_index: chex.Array, num_slots: chex.Array) -> EnvState:
    """Update link-slot and link-slot departure arrays.
    Times are set to negative until turned positive by finalisation (after checks).

    Args:
        state (State): current state
        path (int): path to implement
        initial_slot_index (int): initial slot index
        num_slots (int): number of slots to implement
    """
    state = state.replace(
        link_slot_array=vmap_update_path_links(state.link_slot_array, path, initial_slot_index, num_slots, 1),
        link_slot_departure_array=vmap_update_path_links(state.link_slot_departure_array, path, initial_slot_index, num_slots, state.current_time+state.holding_time)
    )
    return state

implement_vone_action(state, action, total_actions, remaining_actions, params)

Implement action to assign nodes (1, 2, or 0 nodes assigned per action) and assign slots and links for lightpath.

Parameters:

Name Type Description Default
state EnvState

current state

 required
action Array

action to implement (node, node, path_slot_action)

 required
total_actions Scalar

total number of actions to implement for current request

 required
remaining_actions Scalar

remaining actions to implement

 required
k

number of paths to consider

 required
N

number of nodes to assign

 required

Returns:

Name Type Description
state

updated state
Source code in xlron/environments/env_funcs.py 
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
@partial(jax.jit, static_argnums=(4,))
def implement_vone_action(
        state: EnvState,
        action: chex.Array,
        total_actions: chex.Scalar,
        remaining_actions: chex.Scalar,
        params: EnvParams,
):
    """Implement action to assign nodes (1, 2, or 0 nodes assigned per action) and assign slots and links for lightpath.

    Args:
        state: current state
        action: action to implement (node, node, path_slot_action)
        total_actions: total number of actions to implement for current request
        remaining_actions: remaining actions to implement
        k: number of paths to consider
        N: number of nodes to assign

    Returns:
        state: updated state
    """
    request = jax.lax.dynamic_slice(state.request_array[0], ((remaining_actions-1)*2, ), (3, ))
    node_request_s = jax.lax.dynamic_slice(request, (2, ), (1, ))
    requested_datarate = jax.lax.dynamic_slice(request, (1,), (1,))
    node_request_d = jax.lax.dynamic_slice(request, (0, ), (1, ))
    nodes = action[::2]
    path_index, initial_slot_index = process_path_action(state, params, action[1])
    path = get_paths(params, nodes)[path_index]
    se = get_paths_se(params, nodes)[path_index] if params.consider_modulation_format else jnp.array([1])
    num_slots = required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband)

    # jax.debug.print("state.request_array {}", state.request_array, ordered=True)
    # jax.debug.print("path {}", path, ordered=True)
    # jax.debug.print("slots {}", jnp.max(jnp.where(path.reshape(-1,1) == 1, state.link_slot_array, jnp.zeros(params.num_links).reshape(-1,1)), axis=0), ordered=True)
    # jax.debug.print("path_index {}", path_index, ordered=True)
    # jax.debug.print("initial_slot_index {}", initial_slot_index, ordered=True)
    # jax.debug.print("requested_datarate {}", requested_datarate, ordered=True)
    # jax.debug.print("request {}", request, ordered=True)
    # jax.debug.print("se {}", se, ordered=True)
    # jax.debug.print("num_slots {}", num_slots, ordered=True)

    n_nodes = jax.lax.cond(
        total_actions == remaining_actions,
        lambda x: 2, lambda x: 1,
        (total_actions, remaining_actions))
    path_action_only_check = path_action_only(state.request_array[1], state.action_counter, remaining_actions)

    state = jax.lax.cond(
        path_action_only_check,
        lambda x: x[0],
        lambda x: implement_node_action(x[0], x[1], x[2], x[3], x[4], n=x[5]),
        (state, nodes[0], nodes[1], node_request_s, node_request_d, n_nodes)
    )

    state = implement_path_action(state, path, initial_slot_index, num_slots)

    return state

init_action_counter()

Initialize action counter. First index is num unique nodes, second index is total steps, final is remaining steps until completion of request. Only used in VONE environments.

Source code in xlron/environments/env_funcs.py 
539
540
541
542
543
544
def init_action_counter():
    """Initialize action counter.
    First index is num unique nodes, second index is total steps, final is remaining steps until completion of request.
    Only used in VONE environments.
    """
    return jnp.zeros(3, dtype=jnp.int32)

init_action_history(params)

Initialize action history

Source code in xlron/environments/env_funcs.py 
574
575
576
577
@partial(jax.jit, static_argnums=(0,))
def init_action_history(params: EnvParams):
    """Initialize action history"""
    return jnp.full(params.max_edges*2+1, -1)

init_active_lightpaths_array(params)

Initialise active lightpath array. Stores path indices of all active paths on the network in a 1 x M array. M is MIN(max_requests, num_links * link_resources / min_slots). min_slots is the minimum number of slots required for a lightpath i.e. max(values_bw)/ slot_size.

Parameters:

Name Type Description Default
params RSAGNModelEnvParams

Environment parameters

 required

Returns: jnp.array: Active path array (default value -1, empty path)

Source code in xlron/environments/env_funcs.py 
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
def init_active_lightpaths_array(params: RSAGNModelEnvParams):
    """Initialise active lightpath array. Stores path indices of all active paths on the network in a 1 x M array.
    M is MIN(max_requests, num_links * link_resources / min_slots).
    min_slots is the minimum number of slots required for a lightpath i.e. max(values_bw)/ slot_size.

    Args:
        params (RSAGNModelEnvParams): Environment parameters
    Returns:
        jnp.array: Active path array (default value -1, empty path)
    """
    total_slots = params.num_links * params.link_resources  # total slots on networks
    min_slots = jnp.max(params.values_bw.val) / params.slot_size  # minimum number of slots required for lightpath
    return jnp.full((min(int(params.max_requests), int(total_slots / min_slots)),), -1)

init_active_lightpaths_array_departure(params)

Initialise active lightpath array. Stores path indices of all active paths on the network in a 1 x M array. M is MIN(max_requests, num_links * link_resources / min_slots). min_slots is the minimum number of slots required for a lightpath i.e. max(values_bw)/ slot_size.

Parameters:

Name Type Description Default
params RSAGNModelEnvParams

Environment parameters

 required

Returns: jnp.array: Active path array (default value -1, empty path)

Source code in xlron/environments/env_funcs.py 
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
def init_active_lightpaths_array_departure(params: RSAGNModelEnvParams):
    """Initialise active lightpath array. Stores path indices of all active paths on the network in a 1 x M array.
    M is MIN(max_requests, num_links * link_resources / min_slots).
    min_slots is the minimum number of slots required for a lightpath i.e. max(values_bw)/ slot_size.

    Args:
        params (RSAGNModelEnvParams): Environment parameters
    Returns:
        jnp.array: Active path array (default value -1, empty path)
    """
    total_slots = params.num_links * params.link_resources  # total slots on networks
    min_slots = jnp.max(params.values_bw.val) / params.slot_size  # minimum number of slots required for lightpath
    return jnp.full((min(int(params.max_requests), int(total_slots / min_slots)),), 0.)

init_active_path_array(params)

Initialise active path array. Stores details of full path utilised by lightpath on each frequency slot. Args: params (EnvParams): Environment parameters Returns: jnp.array: Active path array

Source code in xlron/environments/env_funcs.py 
2569
2570
2571
2572
2573
2574
2575
2576
def init_active_path_array(params: EnvParams):
    """Initialise active path array. Stores details of full path utilised by lightpath on each frequency slot.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Active path array
    """
    return jnp.full((params.num_links, params.link_resources, params.num_links), 0.)

init_channel_centre_bw_array(params)

Initialise channel centre array. Args: params (EnvParams): Environment parameters Returns: jnp.array: Channel centre array

Source code in xlron/environments/env_funcs.py 
2549
2550
2551
2552
2553
2554
2555
2556
def init_channel_centre_bw_array(params: EnvParams):
    """Initialise channel centre array.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Channel centre array
    """
    return jnp.full((params.num_links, params.link_resources), 0.)

init_channel_power_array(params)

Initialise channel power array.

Parameters:

Name Type Description Default
params EnvParams

Environment parameters

 required

Returns: jnp.array: Channel power array

Source code in xlron/environments/env_funcs.py 
2538
2539
2540
2541
2542
2543
2544
2545
2546
def init_channel_power_array(params: EnvParams):
    """Initialise channel power array.

    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Channel power array
    """
    return jnp.full((params.num_links, params.link_resources), 0.)

init_graph_tuple(state, params, adj, exclude_source_dest=False)

Initialise graph tuple for use with Jraph GNNs. Args: state (EnvState): Environment state params (EnvParams): Environment parameters adj (jnp.array): Adjacency matrix of the graph Returns: jraph.GraphsTuple: Graph tuple

Source code in xlron/environments/env_funcs.py 
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
@partial(jax.jit, static_argnums=(1, 3))
def init_graph_tuple(state: EnvState, params: EnvParams, adj: jnp.array, exclude_source_dest: bool=False) -> jraph.GraphsTuple:
    """Initialise graph tuple for use with Jraph GNNs.
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
        adj (jnp.array): Adjacency matrix of the graph
    Returns:
        jraph.GraphsTuple: Graph tuple
    """
    senders = params.edges.val.T[0]
    receivers = params.edges.val.T[1]

    if exclude_source_dest:
        source_dest_features = jnp.zeros((params.num_nodes, 2))
    else:
        # Get source and dest from request array
        source_dest, _ = read_rsa_request(state.request_array)
        source, dest = source_dest[0], source_dest[2]
        # One-hot encode source and destination (2 additional features)
        source_dest_features = jnp.zeros((params.num_nodes, 2))
        source_dest_features = source_dest_features.at[source.astype(jnp.int32), 0].set(1)
        source_dest_features = source_dest_features.at[dest.astype(jnp.int32), 1].set(-1)

    spectral_features = get_spectral_features(adj, num_features=3)

    if params.__class__.__name__ in ["RSAGNModelEnvParams", "RMSAGNModelEnvParams"]:
        # Normalize by max parameters (converted to linear units)
        max_power = isrs_gn_model.from_dbm(params.max_power)
        normalized_power = jnp.round(state.channel_power_array / max_power, 3)
        max_snr = isrs_gn_model.from_db(params.max_snr)
        normalized_snr = jnp.round(state.link_snr_array / max_snr, 3)
        edge_features = jnp.stack([normalized_snr, normalized_power], axis=-1)
        node_features = jnp.concatenate([spectral_features, source_dest_features], axis=-1)
    elif params.__class__.__name__ == "VONEEnvParams":
        edge_features = state.link_slot_array  # [n_edges] or [n_edges, ...]
        node_features = getattr(state, "node_capacity_array", jnp.zeros(params.num_nodes))
        node_features = node_features.reshape(-1, 1)
        node_features = jnp.concatenate([node_features, spectral_features, source_dest_features], axis=-1)
    else:
        edge_features = state.link_slot_array  # [n_edges] or [n_edges, ...]
        node_features = jnp.concatenate([spectral_features, source_dest_features], axis=-1)

    if params.disable_node_features:
        node_features = jnp.zeros((1,))

    # Handle undirected graphs (duplicate edges after normalization)
    if not params.directed_graph:
        senders_ = jnp.concatenate([senders, receivers])
        receivers = jnp.concatenate([receivers, senders])
        senders = senders_
        edge_features = jnp.repeat(edge_features, 2, axis=0)

    return jraph.GraphsTuple(
        nodes=node_features,
        edges=edge_features,
        senders=senders,
        receivers=receivers,
        n_node=jnp.reshape(params.num_nodes, (1,)),
        n_edge=jnp.reshape(len(senders), (1,)),
        globals=jnp.reshape(state.request_array, (1, -1)),
    )

Initialise link capacity array. Represents available data rate for lightpath on each link. Default is high value (1e6) for unoccupied slots. Once lightpath established, capacity is determined by corresponding entry in path capacity array.

Source code in xlron/environments/env_funcs.py 
2170
2171
2172
2173
2174
def init_link_capacity_array(params):
    """Initialise link capacity array. Represents available data rate for lightpath on each link.
    Default is high value (1e6) for unoccupied slots. Once lightpath established, capacity is determined by
    corresponding entry in path capacity array."""
    return jnp.full((params.num_links, params.link_resources), 1e6)

Initialise link length array. Args: graph (nx.Graph): NetworkX graph Returns:

Source code in xlron/environments/env_funcs.py 
185
186
187
188
189
190
191
192
193
194
195
def init_link_length_array(graph: nx.Graph) -> chex.Array:
    """Initialise link length array.
    Args:
        graph (nx.Graph): NetworkX graph
    Returns:

    """
    link_lengths = []
    for edge in sorted(graph.edges):
        link_lengths.append(graph.edges[edge]["weight"])
    return jnp.array(link_lengths)

Initialise link length array for environements that use GN model of physical layer. We assume each link has spans of equal length.

Parameters:

Name Type Description Default
graph Graph

NetworkX graph

 required

Returns: jnp.array: Link length array (L x max_spans)

Source code in xlron/environments/env_funcs.py 
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
def init_link_length_array_gn_model(graph: nx.Graph, max_span_length: int,  max_spans: int) -> chex.Array:
    """Initialise link length array for environements that use GN model of physical layer.
    We assume each link has spans of equal length.

    Args:
        graph (nx.Graph): NetworkX graph
    Returns:
        jnp.array: Link length array (L x max_spans)
    """
    link_lengths = []
    directed = graph.is_directed()
    graph = graph.to_undirected()
    edges = sorted(graph.edges)
    for edge in edges:
        link_lengths.append(graph.edges[edge]["weight"])
    if directed:
        for edge in edges:
            link_lengths.append(graph.edges[edge]["weight"])
    span_length_array = []
    for length in link_lengths:
        num_spans = math.ceil(length / max_span_length)
        avg_span_length = length / num_spans
        span_lengths = [avg_span_length] * num_spans
        span_lengths.extend([0] * (max_spans - num_spans))
        span_length_array.append(span_lengths)
    return jnp.array(span_length_array)

Initialize empty (all zeroes) link-slot array. 0 means slot is free, -1 means occupied. Args: params (EnvParams): Environment parameters Returns: jnp.array: Link slot array (E x S) where E is number of edges and S is number of slots

Source code in xlron/environments/env_funcs.py 
500
501
502
503
504
505
506
507
@partial(jax.jit, static_argnums=(0,))
def init_link_slot_array(params: EnvParams):
    """Initialize empty (all zeroes) link-slot array. 0 means slot is free, -1 means occupied.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Link slot array (E x S) where E is number of edges and S is number of slots"""
    return jnp.zeros((params.num_links, params.link_resources))

Initialize link mask

Source code in xlron/environments/env_funcs.py 
527
528
529
530
@partial(jax.jit, static_argnums=(0, 1))
def init_link_slot_mask(params: EnvParams, agg: int = 1):
    """Initialize link mask"""
    return jnp.ones(params.k_paths*math.ceil(params.link_resources / agg))

Initialise signal-to-noise ratio (SNR) array. Args: params (EnvParams): Environment parameters Returns: jnp.array: SNR array

Source code in xlron/environments/env_funcs.py 
2527
2528
2529
2530
2531
2532
2533
2534
2535
def init_link_snr_array(params: EnvParams):
    """Initialise signal-to-noise ratio (SNR) array.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: SNR array
    """
    # The SNR is kept in linear units to allow summation of 1/SNR across links
    return jnp.full((params.num_links, params.link_resources), -1e5)

init_mod_format_mask(params)

Initialize link mask

Source code in xlron/environments/env_funcs.py 
533
534
535
536
@partial(jax.jit, static_argnums=(0,))
def init_mod_format_mask(params: EnvParams):
    """Initialize link mask"""
    return jnp.full((params.k_paths*params.link_resources,), -1.0)

init_modulation_format_index_array(params)

Initialise modulation format index array. Args: params (EnvParams): Environment parameters Returns: jnp.array: Modulation format index array

Source code in xlron/environments/env_funcs.py 
2559
2560
2561
2562
2563
2564
2565
2566
def init_modulation_format_index_array(params: EnvParams):
    """Initialise modulation format index array.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Modulation format index array
    """
    return jnp.full((params.num_links, params.link_resources), -1)  # -1 so that highest order is assumed (closest to Gaussian)

init_modulations_array(modulations_filepath=None)

Initialise array of maximum spectral efficiency for modulation format on path.

Parameters:

Name Type Description Default
modulations_filepath str

Path to CSV file containing modulation formats. Defaults to None.

 None

Returns: jnp.array: Array of maximum spectral efficiency for modulation format on path. First two columns are maximum path length and spectral efficiency.

Source code in xlron/environments/env_funcs.py 
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
def init_modulations_array(modulations_filepath: str = None):
    """Initialise array of maximum spectral efficiency for modulation format on path.

    Args:
        modulations_filepath (str, optional): Path to CSV file containing modulation formats. Defaults to None.
    Returns:
        jnp.array: Array of maximum spectral efficiency for modulation format on path.
        First two columns are maximum path length and spectral efficiency.
    """
    f = pathlib.Path(modulations_filepath) if modulations_filepath else (
            pathlib.Path(__file__).parents[1].absolute() / "data" / "modulations" / "modulations.csv")
    modulations = np.genfromtxt(f, delimiter=',')
    # Drop empty first row (headers) and column (name)
    modulations = modulations[1:, 1:]
    return jnp.array(modulations)

init_node_capacity_array(params)

Initialize node array with uniform resources. Args: params (EnvParams): Environment parameters Returns: jnp.array: Node capacity array (N x 1) where N is number of nodes

Source code in xlron/environments/env_funcs.py 
490
491
492
493
494
495
496
497
@partial(jax.jit, static_argnums=(0,))
def init_node_capacity_array(params: EnvParams):
    """Initialize node array with uniform resources.
    Args:
        params (EnvParams): Environment parameters
    Returns:
        jnp.array: Node capacity array (N x 1) where N is number of nodes"""
    return jnp.array([params.node_resources] * params.num_nodes, dtype=jnp.float32)

init_node_mask(params)

Initialize node mask

Source code in xlron/environments/env_funcs.py 
521
522
523
524
@partial(jax.jit, static_argnums=(0,))
def init_node_mask(params: EnvParams):
    """Initialize node mask"""
    return jnp.ones(params.num_nodes)

init_node_resource_array(params)

Array to track node resources occupied by virtual nodes

Source code in xlron/environments/env_funcs.py 
568
569
570
571
@partial(jax.jit, static_argnums=(0,))
def init_node_resource_array(params: EnvParams):
    """Array to track node resources occupied by virtual nodes"""
    return jnp.zeros((params.num_nodes, params.node_resources), dtype=jnp.float32)

init_path_capacity_array(link_length_array, path_link_array, min_request=1, scale_factor=1.0, alpha=0.0002, NF=4.5, B=10000000000000.0, R_s=100000000000.0, beta_2=-2.17e-26, gamma=0.0012, L_s=100000.0, lambda0=1.55e-06)

Calculated from Nevin paper: https://api.repository.cam.ac.uk/server/api/core/bitstreams/b80e7a9c-a86b-4b30-a6d6-05017c60b0c8/content

Parameters:

Name Type Description Default
link_length_array Array

Array of link lengths

 required
path_link_array Array

Array of links on paths

 required
min_request int

Minimum data rate request size. Defaults to 100 GBps.

 1
scale_factor float

Scale factor for link capacity. Defaults to 1.0.

 1.0
alpha float

Fibre attenuation coefficient. Defaults to 0.2e-3 /m

 0.0002
NF float

Amplifier noise figure. Defaults to 4.5 dB.

 4.5
B float

Total modulated bandwidth. Defaults to 10e12 Hz.

 10000000000000.0
R_s float

Symbol rate. Defaults to 100e9 Baud.

 100000000000.0
beta_2 float

Dispersion parameter. Defaults to -21.7e-27 s^2/m.

 -2.17e-26
gamma float

Nonlinear coefficient. Defaults to 1.2e-3 /W/m.

 0.0012
L_s float

Span length. Defaults to 100e3 m.

 100000.0
lambda0 float

Wavelength. Defaults to 1550e-9 m.

 1.55e-06

Returns:

Type Description
Array

chex.Array: Array of link capacities in Gbps
Source code in xlron/environments/env_funcs.py 
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
def init_path_capacity_array(
        link_length_array: chex.Array,
        path_link_array: chex.Array,
        min_request=1,  # Minimum data rate request size
        scale_factor=1.0,  # Scale factor for link capacity
        alpha=0.2e-3,  # Fibre attenuation coefficient
        NF=4.5,  # Amplifier noise figure
        B=10e12,  # Total modulated bandwidth
        R_s=100e9,  # Symbol rate
        beta_2=-21.7e-27,  # Dispersion parameter
        gamma=1.2e-3,  # Nonlinear coefficient
        L_s=100e3,  # span length
        lambda0=1550e-9,  # Wavelength
) -> chex.Array:
    """Calculated from Nevin paper:
    https://api.repository.cam.ac.uk/server/api/core/bitstreams/b80e7a9c-a86b-4b30-a6d6-05017c60b0c8/content

    Args:
        link_length_array (chex.Array): Array of link lengths
        path_link_array (chex.Array): Array of links on paths
        min_request (int, optional): Minimum data rate request size. Defaults to 100 GBps.
        scale_factor (float, optional): Scale factor for link capacity. Defaults to 1.0.
        alpha (float, optional): Fibre attenuation coefficient. Defaults to 0.2e-3 /m
        NF (float, optional): Amplifier noise figure. Defaults to 4.5 dB.
        B (float, optional): Total modulated bandwidth. Defaults to 10e12 Hz.
        R_s (float, optional): Symbol rate. Defaults to 100e9 Baud.
        beta_2 (float, optional): Dispersion parameter. Defaults to -21.7e-27 s^2/m.
        gamma (float, optional): Nonlinear coefficient. Defaults to 1.2e-3 /W/m.
        L_s (float, optional): Span length. Defaults to 100e3 m.
        lambda0 (float, optional): Wavelength. Defaults to 1550e-9 m.

    Returns:
        chex.Array: Array of link capacities in Gbps
    """
    path_length_array = jnp.dot(path_link_array, link_length_array)
    path_capacity_array = calculate_path_capacity(
        path_length_array,
        min_request=min_request,
        scale_factor=scale_factor,
        alpha=alpha,
        NF=NF,
        B=B,
        R_s=R_s,
        beta_2=beta_2,
        gamma=gamma,
        L_s=L_s,
        lambda0=lambda0,
    )
    return path_capacity_array

init_path_index_array(params)

Initialise path index array. Represents index of lightpath occupying each slot.

Source code in xlron/environments/env_funcs.py 
2177
2178
2179
def init_path_index_array(params):
    """Initialise path index array. Represents index of lightpath occupying each slot."""
    return jnp.full((params.num_links, params.link_resources), -1)

init_path_length_array(path_link_array, graph)

Initialise path length array.

Parameters:

Name Type Description Default
path_link_array Array

Path-link array

 required
graph Graph

NetworkX graph

 required

Returns: chex.Array: Path length array

Source code in xlron/environments/env_funcs.py 
331
332
333
334
335
336
337
338
339
340
341
342
def init_path_length_array(path_link_array: chex.Array, graph: nx.Graph) -> chex.Array:
    """Initialise path length array.

    Args:
        path_link_array (chex.Array): Path-link array
        graph (nx.Graph): NetworkX graph
    Returns:
        chex.Array: Path length array
    """
    link_length_array = init_link_length_array(graph)
    path_lengths = jnp.dot(path_link_array, link_length_array)
    return path_lengths

Initialise path-link array. Each path is defined by a link utilisation array (one row in the path-link array). 1 indicates link corresponding to index is used, 0 indicates not used.

Parameters:

Name Type Description Default
graph Graph

NetworkX graph

 required
k int

Number of paths

 required
disjoint bool

Whether to use edge-disjoint paths. Defaults to False.

 False
weight str

Sort paths by edge attribute. Defaults to "weight".

 'weight'
directed bool

Whether graph is directed. Defaults to False.

 False
modulations_array Array

Array of maximum spectral efficiency for modulation format on path. Defaults to None.

 None
rwa_lr bool

Whether the environment is RWA with lightpath reuse (affects path ordering).

 False
path_snr bool

If GN model is used, include extra row of zeroes for unutilised paths

 False

Returns:

Type Description
Array

chex.Array: Path-link array (N(N-1)*k x E) where N is number of nodes, E is number of edges, k is number of shortest paths
Source code in xlron/environments/env_funcs.py 
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
def init_path_link_array(
        graph: nx.Graph,
        k: int,
        disjoint: bool = False,
        weight: str = "weight",
        directed: bool = False,
        modulations_array: chex.Array = None,
        rwa_lr: bool = False,
        scale_factor: float = 1.0,
        path_snr: bool = False,
) -> chex.Array:
    """Initialise path-link array.
    Each path is defined by a link utilisation array (one row in the path-link array).
    1 indicates link corresponding to index is used, 0 indicates not used.

    Args:
        graph (nx.Graph): NetworkX graph
        k (int): Number of paths
        disjoint (bool, optional): Whether to use edge-disjoint paths. Defaults to False.
        weight (str, optional): Sort paths by edge attribute. Defaults to "weight".
        directed (bool, optional): Whether graph is directed. Defaults to False.
        modulations_array (chex.Array, optional): Array of maximum spectral efficiency for modulation format on path. Defaults to None.
        rwa_lr (bool, optional): Whether the environment is RWA with lightpath reuse (affects path ordering).
        path_snr (bool, optional): If GN model is used, include extra row of zeroes for unutilised paths
        to ensure correct SNR calculation for empty paths (path index -1).

    Returns:
        chex.Array: Path-link array (N(N-1)*k x E) where N is number of nodes, E is number of edges, k is number of shortest paths
    """
    def get_k_shortest_paths(g, source, target, k, weight):
        return list(
            islice(nx.shortest_simple_paths(g, source, target, weight=weight), k)
        )

    def get_k_disjoint_shortest_paths(g, source, target, k, weight):
        k_paths_disjoint_unsorted = list(nx.edge_disjoint_paths(g, source, target))
        k_paths_shortest = get_k_shortest_paths(g, source, target, k, weight=weight)

        # Keep disjoint paths and add unique shortest paths until k paths reached
        disjoint_ids = [tuple(path) for path in k_paths_disjoint_unsorted]
        k_paths = k_paths_disjoint_unsorted
        for path in k_paths_shortest:
            if tuple(path) not in disjoint_ids:
                k_paths.append(path)
        k_paths = k_paths[:k]
        return k_paths

    paths = []
    edges = sorted(graph.edges)

    # Get the k-shortest paths for each node pair
    k_path_collections = []
    get_paths = get_k_disjoint_shortest_paths if disjoint else get_k_shortest_paths
    for node_pair in combinations(graph.nodes, 2):

        k_paths = get_paths(graph, node_pair[0], node_pair[1], k, weight=weight)
        k_path_collections.append(k_paths)

    if directed:  # Get paths in reverse direction
        for node_pair in combinations(graph.nodes, 2):
            k_paths_rev = get_paths(graph, node_pair[1], node_pair[0], k, weight=weight)
            k_path_collections.append(k_paths_rev)

    # Sort the paths for each node pair
    max_missing_paths = 0
    for k_paths in k_path_collections:

        source, dest = k_paths[0][0], k_paths[0][-1]

        # Sort the paths by # of hops then by length, or just length
        path_lengths = [nx.path_weight(graph, path, weight='weight') for path in k_paths]
        path_num_links = [len(path) - 1 for path in k_paths]

        # Get maximum spectral efficiency for modulation format on path
        if modulations_array is not None and rwa_lr is not True:
            se_of_path = []
            modulations_array = modulations_array[::-1]
            for length in path_lengths:
                for modulation in modulations_array:
                    if length <= modulation[0]:
                        se_of_path.append(modulation[1])
                        break
            # Sorting by the num_links/se instead of just path length is observed to improve performance
            path_weighting = [num_links/se for se, num_links in zip(se_of_path, path_num_links)]
        elif rwa_lr:
            path_capacity = [float(calculate_path_capacity(path_length, scale_factor=scale_factor)) for path_length in path_lengths]
            path_weighting = [num_links/path_capacity for num_links, path_capacity in zip(path_num_links, path_capacity)]
        elif weight is None:
            path_weighting = path_num_links
        else:
            path_weighting = path_lengths

        # if less then k unique paths, add empty paths
        empty_path = [0] * len(graph.edges)
        num_missing_paths = k - len(k_paths)
        max_missing_paths = max(max_missing_paths, num_missing_paths)
        k_paths = k_paths + [empty_path] * num_missing_paths
        path_weighting = path_weighting + [1e6] * num_missing_paths
        path_lengths = path_lengths + [1e6] * num_missing_paths

        # Sort by number of links then by length (or just by length if weight is specified)
        unsorted_paths = zip(k_paths, path_weighting, path_lengths)
        k_paths_sorted = [(source, dest, weighting, path) for path, weighting, _ in sorted(unsorted_paths, key=lambda x: (x[1], 1/x[2]) if weight is None else x[2])]

        # Keep only first k paths
        k_paths_sorted = k_paths_sorted[:k]

        prev_link_usage = empty_path
        for k_path in k_paths_sorted:
            k_path = k_path[-1]
            link_usage = [0]*len(graph.edges)  # Initialise empty path
            if sum(k_path) == 0:
                link_usage = prev_link_usage
            else:
                for i in range(len(k_path)-1):
                    s, d = k_path[i], k_path[i + 1]
                    for edge_index, edge in enumerate(edges):
                        condition = (edge[0] == s and edge[1] == d) if directed else \
                            ((edge[0] == s and edge[1] == d) or (edge[0] == d and edge[1] == s))
                        if condition:
                            link_usage[edge_index] = 1
            path = link_usage
            prev_link_usage = link_usage
            paths.append(path)

    # If using GN model, add extra row of zeroes for empty paths for SNR calculation
    if path_snr:
        empty_path = [0] * len(graph.edges)
        paths.append(empty_path)

    return jnp.array(paths, dtype=jnp.float32)

init_path_se_array(path_length_array, modulations_array)

Initialise array of maximum spectral efficiency for highest-order modulation format on path.

Parameters:

Name Type Description Default
path_length_array array

Array of path lengths

 required
modulations_array array

Array of maximum spectral efficiency for modulation format on path

 required

Returns:

Type Description

jnp.array: Array of maximum spectral efficiency for on path
Source code in xlron/environments/env_funcs.py 
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
def init_path_se_array(path_length_array, modulations_array):
    """Initialise array of maximum spectral efficiency for highest-order modulation format on path.

    Args:
        path_length_array (jnp.array): Array of path lengths
        modulations_array (jnp.array): Array of maximum spectral efficiency for modulation format on path

    Returns:
        jnp.array: Array of maximum spectral efficiency for on path
    """
    se_list = []
    # Flip the modulation array so that the shortest path length is first
    modulations_array = modulations_array[::-1]
    for length in path_length_array:
        for modulation in modulations_array:
            if length <= modulation[0]:
                se_list.append(modulation[1])
                break
    return jnp.array(se_list)

init_rsa_request_array()

Initialize request array

Source code in xlron/environments/env_funcs.py 
516
517
518
def init_rsa_request_array():
    """Initialize request array"""
    return jnp.zeros(3)

init_traffic_matrix(key, params)

Initialize traffic matrix. Allows for random traffic matrix or uniform traffic matrix. Source-dest traffic requests are sampled probabilistically from the resulting traffic matrix.

Parameters:

Name Type Description Default
key PRNGKey

PRNG key

 required
params EnvParams

Environment parameters

 required

Returns:

Type Description

jnp.array: Traffic matrix
Source code in xlron/environments/env_funcs.py 
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
@partial(jax.jit, static_argnums=(1,))
def init_traffic_matrix(key: chex.PRNGKey, params: EnvParams):
    """Initialize traffic matrix. Allows for random traffic matrix or uniform traffic matrix.
    Source-dest traffic requests are sampled probabilistically from the resulting traffic matrix.

    Args:
        key (chex.PRNGKey): PRNG key
        params (EnvParams): Environment parameters

    Returns:
        jnp.array: Traffic matrix
    """
    if params.random_traffic:
        traffic_matrix = jax.random.uniform(key, shape=(params.num_nodes, params.num_nodes))
    else:
        traffic_matrix = jnp.ones((params.num_nodes, params.num_nodes))
    diag_elements = jnp.diag_indices_from(traffic_matrix)
    # Set main diagonal to zero so no requests from node to itself
    traffic_matrix = traffic_matrix.at[diag_elements].set(0)
    traffic_matrix = normalise_traffic_matrix(traffic_matrix)
    return traffic_matrix

init_virtual_topology_patterns(pattern_names)

Initialise virtual topology patterns. First 3 digits comprise the "action counter": first index is num unique nodes, second index is total steps, final is remaining steps until completion of request. Remaining digits define the topology pattern, with 1 to indicate links and other positive integers are node indices.

Parameters:

Name Type Description Default
pattern_names list

List of virtual topology pattern names

 required

Returns:

Type Description
Array

chex.Array: Array of virtual topology patterns
Source code in xlron/environments/env_funcs.py 
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
def init_virtual_topology_patterns(pattern_names: str) -> chex.Array:
    """Initialise virtual topology patterns.
    First 3 digits comprise the "action counter": first index is num unique nodes, second index is total steps,
    final is remaining steps until completion of request.
    Remaining digits define the topology pattern, with 1 to indicate links and other positive integers are node indices.

    Args:
        pattern_names (list): List of virtual topology pattern names

    Returns:
        chex.Array: Array of virtual topology patterns
    """
    patterns = []
    # TODO - Allow 2 node requests in VONE (check if any modifications necessary other than below)
    #if "2_bus" in pattern_names:
    #    patterns.append([2, 1, 1, 2, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0])
    if "3_bus" in pattern_names:
        patterns.append([3, 2, 2, 2, 1, 3, 1, 4])
    if "3_ring" in pattern_names:
        patterns.append([3, 3, 3, 2, 1, 3, 1, 4, 1, 2])
    if "4_bus" in pattern_names:
        patterns.append([4, 3, 3, 2, 1, 3, 1, 4, 1, 5])
    if "4_ring" in pattern_names:
        patterns.append([4, 4, 4, 2, 1, 3, 1, 4, 1, 5, 1, 2])
    if "5_bus" in pattern_names:
        patterns.append([5, 4, 4, 2, 1, 3, 1, 4, 1, 5, 1, 6])
    if "5_ring" in pattern_names:
        patterns.append([5, 5, 5, 2, 1, 3, 1, 4, 1, 5, 1, 6, 1, 2])
    if "6_bus" in pattern_names:
        patterns.append([6, 5, 5, 2, 1, 3, 1, 4, 1, 5, 1, 6, 1, 7])
    max_length = max([len(pattern) for pattern in patterns])
    # Pad patterns with zeroes to match longest
    for pattern in patterns:
        pattern.extend([0]*(max_length-len(pattern)))
    return jnp.array(patterns, dtype=jnp.int32)

init_vone_request_array(params)

Initialize request array either with uniform resources

Source code in xlron/environments/env_funcs.py 
510
511
512
513
@partial(jax.jit, static_argnums=(0,))
def init_vone_request_array(params: EnvParams):
    """Initialize request array either with uniform resources"""
    return jnp.zeros((2, params.max_edges*2+1, ))

kca_ff(state, params)

Congestion-aware First Fit. Only suitable for RSA/RMSA. Method:

Source code in xlron/heuristics/heuristics.py 
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
@partial(jax.jit, static_argnums=(1,))
def kca_ff(state: EnvState, params: EnvParams) -> chex.Array:
    """Congestion-aware First Fit. Only suitable for RSA/RMSA.
    Method:

    """
    mask = get_action_mask(state, params)
    # Get index of first available slots for each path
    first_slots = first_fit(state, params)
    # Get nodes
    nodes_sd, _ = read_rsa_request(state.request_array)
    # Initialise array to hold congestion on each path
    path_congestion_array = jnp.full((mask.shape[0],), 0.)
    link_weights = get_link_weights(state, params)

    def get_path_congestion(i, val):
        # Get links on path
        path = get_paths(params, nodes_sd)[i]
        # Get congestion
        path_link_congestion = jnp.multiply(link_weights, path)
        path_congestion = jnp.sum(path_link_congestion).reshape((1,))
        return jax.lax.dynamic_update_slice(val, path_congestion, (i,))

    path_congestion_array = jax.lax.fori_loop(0, mask.shape[0], get_path_congestion, path_congestion_array)
    path_index = jnp.argmin(path_congestion_array)
    slot_index = first_slots[path_index] % params.link_resources
    action = path_index * params.link_resources + slot_index
    return action

kmc_ff(state, params)

K-Minimum Cut. Only suitable for RSA/RMSA. Method: 1. Go through action mask and find the first available slot on all paths. 2. For each path, allocate the first available slot. 3. Sum number of new consecutive zero regions (cuts) created by assignment (on each link) 4. Choose path that creates the fewest cuts.

Source code in xlron/heuristics/heuristics.py 
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
@partial(jax.jit, static_argnums=(1,))
def kmc_ff(state: EnvState, params: EnvParams) -> chex.Array:
    """K-Minimum Cut. Only suitable for RSA/RMSA.
    Method:
    1. Go through action mask and find the first available slot on all paths.
    2. For each path, allocate the first available slot.
    3. Sum number of new consecutive zero regions (cuts) created by assignment (on each link)
    4. Choose path that creates the fewest cuts.
    """
    mask = get_action_mask(state, params)
    first_slots = first_fit(state, params)
    link_slot_array = jnp.where(state.link_slot_array < 0, 1., state.link_slot_array)
    nodes_sd, requested_bw = read_rsa_request(state.request_array)
    block_sizes = jax.vmap(find_block_sizes, in_axes=(0,))(link_slot_array)
    block_sizes_mask = jnp.where(block_sizes > 0, 1, 0.)  # Binary array showing initial block starts
    block_count = jnp.sum(block_sizes_mask, axis=1)

    def get_cuts_on_path(i, result):
        initial_slot_index = first_slots[i] % params.link_resources
        path = get_paths(params, nodes_sd)[i]
        se = get_paths_se(params, nodes_sd)[i] if params.consider_modulation_format else 1
        num_slots = required_slots(requested_bw, se, params.slot_size, guardband=params.guardband)
        # Make link-slot_array positive
        updated_slots = vmap_set_path_links(link_slot_array, path, initial_slot_index, num_slots, 1.)
        updated_block_sizes = jax.vmap(find_block_sizes, in_axes=(0,))(updated_slots)
        updated_block_sizes_mask = jnp.where(updated_block_sizes > 0, 1, 0)  # Binary array showing updated block starts
        updated_block_count = jnp.sum(updated_block_sizes_mask, axis=1)
        num_cuts = jax.lax.cond(
            mask[i][initial_slot_index] == 0.,  # If true, no valid action for path
            lambda x: jnp.full((1,), params.link_resources*params.num_links).astype(jnp.float32),  # Return max no. of cuts
            lambda x: jnp.sum(jnp.maximum(updated_block_count - block_count, 0.)).reshape((1,)),  # Else, return number of cuts
            1.
        )
        result = jax.lax.dynamic_update_slice(result, num_cuts, (i,))
        return result

    # Initialise array to hold number of cuts on each path
    path_cuts_array = jnp.full((mask.shape[0],), 0.)
    path_cuts_array = jax.lax.fori_loop(0, mask.shape[0], get_cuts_on_path, path_cuts_array)
    path_index = jnp.argmin(path_cuts_array)
    slot_index = first_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

kme_ff(state, params)

K-Minimum Entropy. Only suitable for RSA/RMSA. Method: 1. Go through action mask and find the first available slot on all paths. 2. For each path, allocate the first available slot. 3. Sum number of new consecutive zero regions (cuts) created by assignment (on each link) 4. Choose path that creates the fewest cuts.

Source code in xlron/heuristics/heuristics.py 
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
@partial(jax.jit, static_argnums=(1,))
def kme_ff(state: EnvState, params: EnvParams) -> chex.Array:
    """K-Minimum Entropy. Only suitable for RSA/RMSA.
    Method:
    1. Go through action mask and find the first available slot on all paths.
    2. For each path, allocate the first available slot.
    3. Sum number of new consecutive zero regions (cuts) created by assignment (on each link)
    4. Choose path that creates the fewest cuts.
    """
    mask = get_action_mask(state, params)
    first_slots = first_fit(state, params)
    link_slot_array = jnp.where(state.link_slot_array < 0, 1., state.link_slot_array)
    nodes_sd, requested_bw = read_rsa_request(state.request_array)
    max_entropy = jnp.sum(jnp.log(params.link_resources)) * params.num_links

    def get_link_entropy(blocks):
        ent = jax.vmap(lambda x: jnp.sum(x/params.link_resources * jnp.log(params.link_resources/x)), in_axes=0)(blocks)
        return jnp.sum(jnp.where(blocks > 0, ent, 0))

    def get_entropy_on_path(i, result):
        initial_slot_index = first_slots[i] % params.link_resources
        path = get_paths(params, nodes_sd)[i]
        se = get_paths_se(params, nodes_sd)[i] if params.consider_modulation_format else 1
        num_slots = required_slots(requested_bw, se, params.slot_size, guardband=params.guardband)
        # Make link-slot_array positive
        updated_slots = vmap_set_path_links(link_slot_array, path, initial_slot_index, num_slots, 1.)
        updated_block_sizes = jax.vmap(find_block_sizes, in_axes=(0,))(updated_slots)
        updated_entropy = jax.vmap(get_link_entropy, in_axes=(0,))(updated_block_sizes)
        new_path_entropy = jnp.sum(jnp.dot(path, updated_entropy)).reshape((1,))
        new_path_entropy = jax.lax.cond(
            mask[i][initial_slot_index] == 0.,  # If true, no valid action for path
            lambda x: max_entropy.astype(jnp.float32).reshape((1,)),  # Return maximum entropy
            lambda x: new_path_entropy,  # Else, return number of cuts
            1.
        )
        result = jax.lax.dynamic_update_slice(result, new_path_entropy, (i,))
        return result

    path_entropy_array = jnp.full((mask.shape[0],), 0.)
    path_entropy_array = jax.lax.fori_loop(0, mask.shape[0], get_entropy_on_path, path_entropy_array)
    path_index = jnp.argmin(path_entropy_array)
    slot_index = first_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

kmf_ff(state, params)

K-Minimum Frag-size. Only suitable for RSA/RMSA. Method: 1. Go through action mask and find the first available slot on all paths. 2. For each path, allocate the first available slot. 3. Sum number of new consecutive zero regions (cuts) created by assignment (on each link) 4. Choose path that creates the fewest cuts.

Source code in xlron/heuristics/heuristics.py 
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
@partial(jax.jit, static_argnums=(1,))
def kmf_ff(state: RSAEnvState, params: RSAEnvParams) -> chex.Array:
    """K-Minimum Frag-size. Only suitable for RSA/RMSA.
    Method:
    1. Go through action mask and find the first available slot on all paths.
    2. For each path, allocate the first available slot.
    3. Sum number of new consecutive zero regions (cuts) created by assignment (on each link)
    4. Choose path that creates the fewest cuts.
    """
    mask = get_action_mask(state, params)
    first_slots = first_fit(state, params)
    link_slot_array = jnp.where(state.link_slot_array < 0, 1., state.link_slot_array)
    nodes_sd, requested_bw = read_rsa_request(state.request_array)
    blocks = jax.vmap(find_block_sizes, in_axes=(0,))(link_slot_array)

    def get_frags_on_path(i, result):
        initial_slot_index = first_slots[i] % params.link_resources
        path = get_paths(params, nodes_sd)[i]
        se = get_paths_se(params, nodes_sd)[i] if params.consider_modulation_format else 1
        num_slots = required_slots(requested_bw, se, params.slot_size, guardband=params.guardband)
        # Mask on path links
        block_sizes = jax.vmap(lambda x, y: jnp.where(x > 0, y, 0.), in_axes=(0, 0))(path, blocks)
        updated_slots = vmap_set_path_links(state.link_slot_array, path, initial_slot_index, num_slots, -1)
        updated_block_sizes = jax.vmap(find_block_sizes, in_axes=(0,))(updated_slots)
        # Mask on path links
        updated_block_sizes = jax.vmap(lambda x, y: jnp.where(x > 0, y, 0.), in_axes=(0, 0))(path, updated_block_sizes)
        difference = updated_block_sizes - block_sizes
        new_frags = jnp.where(difference != 0, block_sizes + difference, 0.)
        # Slice new frags up to initial slot index (so as to only consider frags to the left)
        new_frags = jnp.where(jnp.arange(params.link_resources) < initial_slot_index, new_frags, 0.)
        new_frag_size = jnp.sum(new_frags)
        num_frags = jax.lax.cond(
            mask[i][initial_slot_index] == 0.,  # If true, no valid action for path
            lambda x: jnp.full((1,), float(params.link_resources * params.num_links)),  # Return max frag size
            lambda x: new_frag_size.reshape((1,)),
            # Else, return number of cuts
            1.
        )
        result = jax.lax.dynamic_update_slice(result, num_frags, (i,))
        return result

    # Initialise array to hold number of cuts on each path
    path_frags_array = jnp.full((mask.shape[0],), 0.)
    path_frags_array = jax.lax.fori_loop(0, mask.shape[0], get_frags_on_path, path_frags_array)
    path_index = jnp.argmin(path_frags_array)
    slot_index = first_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

ksp_bf(state, params)

Get the first available slot from all k-shortest paths Method: Go through action mask and find the first available slot, starting from shortest path

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required

Returns:

Type Description
Array

chex.Array: Action
Source code in xlron/heuristics/heuristics.py 
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
@partial(jax.jit, static_argnums=(1,))
def ksp_bf(state: EnvState, params: EnvParams) -> chex.Array:
    """Get the first available slot from all k-shortest paths
    Method: Go through action mask and find the first available slot, starting from shortest path

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters

    Returns:
        chex.Array: Action
    """
    best_slots, fitness = best_fit(state, params)
    # Chosen path is the first one with an available slot
    path_index = jnp.argmin(jnp.where(fitness < jnp.inf, 0, 1))
    slot_index = best_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

ksp_ff(state, params)

Get the first available slot from the shortest available path Method: Go through action mask and find the first available slot, starting from shortest path

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required

Returns:

Type Description
Array

chex.Array: Action
Source code in xlron/heuristics/heuristics.py 
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
@partial(jax.jit, static_argnums=(1,))
def ksp_ff(state: EnvState, params: EnvParams) -> chex.Array:
    """Get the first available slot from the shortest available path
    Method: Go through action mask and find the first available slot, starting from shortest path

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters

    Returns:
        chex.Array: Action
    """
    first_slots = first_fit(state, params)
    # Chosen path is the first one with an available slot
    path_index = jnp.argmax(first_slots < params.link_resources)
    slot_index = first_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

ksp_ff_multiband(state, params)

Get the first available slot from all k-shortest paths in multiband scenario Method: Go through action mask and find the first available slot, starting from shortest path

Parameters:

Name Type Description Default
state MultiBandRSAEnvState

Environment state specific to multiband operations

 required
params MultiBandRSAEnvParams

Environment parameters including multiband details

 required

Returns: chex.Array: Action

Source code in xlron/heuristics/heuristics.py 
30
31
32
33
34
35
36
37
38
39
40
def ksp_ff_multiband(state: EnvState, params: EnvParams) -> chex.Array:
    """Get the first available slot from all k-shortest paths in multiband scenario
    Method: Go through action mask and find the first available slot, starting from shortest path

    Args:
        state (MultiBandRSAEnvState): Environment state specific to multiband operations
        params (MultiBandRSAEnvParams): Environment parameters including multiband details
    Returns:
        chex.Array: Action
    """
    pass

ksp_lf(state, params)

Get the last available slot on the shortest available path Method: Go through action mask and find the last available slot, starting from shortest path

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required

Returns:

Type Description
Array

chex.Array: Action
Source code in xlron/heuristics/heuristics.py 
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
@partial(jax.jit, static_argnums=(1,))
def ksp_lf(state: EnvState, params: EnvParams) -> chex.Array:
    """Get the last available slot on the shortest available path
    Method: Go through action mask and find the last available slot, starting from shortest path

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters

    Returns:
        chex.Array: Action
    """
    last_slots = last_fit(state, params)
    # Chosen path is the first one with an available slot
    path_index = jnp.argmax(last_slots < params.link_resources)
    slot_index = last_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

ksp_mu(state, params, unique_lightpaths, relative)

Get the most-used slot on the shortest available path. Method: Go through action mask and find the utilisation of available slots on each path. Find the shortest available path and choose the most utilised slot on that path.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required
unique_lightpaths bool

Whether to consider unique lightpaths

 required
relative bool

Whether to return relative utilisation

 required

Returns:

Type Description
Array

chex.Array: Action
Source code in xlron/heuristics/heuristics.py 
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
@partial(jax.jit, static_argnums=(1, 2, 3))
def ksp_mu(state: EnvState, params: EnvParams, unique_lightpaths: bool, relative: bool) -> chex.Array:
    """Get the most-used slot on the shortest available path.
    Method: Go through action mask and find the utilisation of available slots on each path.
    Find the shortest available path and choose the most utilised slot on that path.

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
        unique_lightpaths (bool): Whether to consider unique lightpaths
        relative (bool): Whether to return relative utilisation

    Returns:
        chex.Array: Action
    """
    mask = get_action_mask(state, params)
    most_used_slots = most_used(state, params, unique_lightpaths, relative)
    # Get usage of available slots
    most_used_mask = most_used_slots * mask
    # Get index of most-used available slot for each path
    most_used_slots = jnp.argmax(most_used_mask, axis=1).astype(jnp.int32)
    # Chosen path is the first one with an available slot
    available_paths = jnp.max(mask, axis=1)
    path_index = jnp.argmax(available_paths)
    slot_index = most_used_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

last_fit(state, params)

Last-Fit Spectrum Allocation. Returns the last fit slot for each path.

Source code in xlron/heuristics/heuristics.py 
476
477
478
479
480
481
482
483
484
485
def last_fit(state: EnvState, params: EnvParams) -> chex.Array:
    """Last-Fit Spectrum Allocation. Returns the last fit slot for each path."""
    mask = get_action_mask(state, params)
    # Add a column of ones to the mask to make sure that occupied paths have non-zero index in "last_slots"
    mask = jnp.concatenate((jnp.full((mask.shape[0], 1), 1), mask), axis=1)
    # Get index of last available slots for each path
    last_slots = jnp.argmax(mask[:, ::-1], axis=1)
    # Convert to index from the left
    last_slots = params.link_resources - last_slots - 1
    return last_slots

lf_ksp(state, params)

Get the last available slot from all paths Method: Go through action mask and find the last available slot on all paths

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required

Returns:

Type Description
Array

chex.Array: Action
Source code in xlron/heuristics/heuristics.py 
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
@partial(jax.jit, static_argnums=(1,))
def lf_ksp(state: EnvState, params: EnvParams) -> chex.Array:
    """Get the last available slot from all paths
    Method: Go through action mask and find the last available slot on all paths

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters

    Returns:
        chex.Array: Action
    """
    last_slots = last_fit(state, params)
    # Chosen path is the one with the highest index of last available slot
    path_index = jnp.argmax(last_slots)
    slot_index = last_slots[path_index] % params.link_resources
    # Convert indices to action
    action = path_index * params.link_resources + slot_index
    return action

make_graph(topology_name='conus', topology_directory=None)

Create graph from topology definition. Topologies must be defined in JSON format in the topologies directory and named as the topology name with .json extension.

Parameters:

Name Type Description Default
topology_name str

topology name

 'conus'
topology_directory str

topology directory

 None

Returns:

Name Type Description
graph

graph
Source code in xlron/environments/env_funcs.py 
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
def make_graph(topology_name: str = "conus", topology_directory: str = None):
    """Create graph from topology definition.
    Topologies must be defined in JSON format in the topologies directory and
    named as the topology name with .json extension.

    Args:
        topology_name: topology name
        topology_directory: topology directory

    Returns:
        graph: graph
    """
    topology_path = pathlib.Path(topology_directory) if topology_directory else (
            pathlib.Path(__file__).parents[1].absolute() / "data" / "topologies")
    # Create topology
    if topology_name == "4node":
        # 4 node ring
        graph = nx.from_numpy_array(np.array([[0, 1, 0, 1],
                                            [1, 0, 1, 0],
                                               [0, 1, 0, 1],
                                               [1, 0, 1, 0]]))
        # Add edge weights to graph
        nx.set_edge_attributes(graph, {(0, 1): 4, (1, 2): 3, (2, 3): 2, (3, 0): 1}, "weight")
    elif topology_name == "7node":
        # 7 node ring
        graph = nx.from_numpy_array(jnp.array([[0, 1, 0, 0, 0, 0, 1],
                                               [1, 0, 1, 0, 0, 0, 0],
                                               [0, 1, 0, 1, 0, 0, 0],
                                               [0, 0, 1, 0, 1, 0, 0],
                                               [0, 0, 0, 1, 0, 1, 0],
                                               [0, 0, 0, 0, 1, 0, 1],
                                               [1, 0, 0, 0, 0, 1, 0]]))
        # Add edge weights to graph
        nx.set_edge_attributes(graph, {(0, 1): 4, (1, 2): 3, (2, 3): 2, (3, 4): 1, (4, 5): 2, (5, 6): 3, (6, 0): 4}, "weight")
    else:
        with open(topology_path / f"{topology_name}.json") as f:
            graph = nx.node_link_graph(json.load(f))
    return graph

mask_nodes(state, num_nodes)

Returns mask of valid actions for node selection. 1 for valid action, 0 for invalid action.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
num_nodes Scalar

Number of nodes

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
@partial(jax.jit, static_argnums=(1,))
def mask_nodes(state: EnvState, num_nodes: chex.Scalar) -> EnvState:
    """Returns mask of valid actions for node selection. 1 for valid action, 0 for invalid action.

    Args:
        state: Environment state
        num_nodes: Number of nodes

    Returns:
        state: Updated environment state
    """
    total_actions = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.action_counter, 1, 1))
    remaining_actions = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.action_counter, 2, 1))
    full_request = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.request_array, 0, 1))
    virtual_topology = jnp.squeeze(jax.lax.dynamic_slice_in_dim(state.request_array, 1, 1))
    request = jax.lax.dynamic_slice_in_dim(full_request, (remaining_actions - 1) * 2, 3)
    node_request_s = jax.lax.dynamic_slice_in_dim(request, 2, 1)
    node_request_d = jax.lax.dynamic_slice_in_dim(request, 0, 1)
    prev_action = jax.lax.dynamic_slice_in_dim(state.action_history, (remaining_actions) * 2, 3)
    prev_dest = jax.lax.dynamic_slice_in_dim(prev_action, 0, 1)
    node_indices = jnp.arange(0, num_nodes)
    # Get requested indices from request array virtual topology
    requested_indices = jax.lax.dynamic_slice_in_dim(virtual_topology, (remaining_actions-1)*2, 3)
    requested_index_d = jax.lax.dynamic_slice_in_dim(requested_indices, 0, 1)
    # Get index of previous selected node
    prev_selected_node = jnp.where(virtual_topology == requested_index_d, state.action_history, jnp.full(virtual_topology.shape, -1))
    # will be current index if node only occurs once in virtual topology or will be different index if occurs more than once
    prev_selected_index = jnp.argmax(prev_selected_node)
    prev_selected_node_d = jax.lax.dynamic_slice_in_dim(state.action_history, prev_selected_index, 1)

    # If first action, source and dest both to be assigned -> just mask all nodes based on resources
    # Thereafter, source must be previous dest. Dest can be any node (except previous allocations).
    state = state.replace(
        node_mask_s=jax.lax.cond(
            jnp.equal(remaining_actions, total_actions),
            lambda x: jnp.where(
                state.node_capacity_array >= node_request_s,
                x,
                jnp.zeros(num_nodes)
            ),
            lambda x: jnp.where(
                node_indices == prev_dest,
                x,
                jnp.zeros(num_nodes)
            ),
            jnp.ones(num_nodes),
        )
    )
    state = state.replace(
        node_mask_d=jnp.where(
            state.node_capacity_array >= node_request_d,
            jnp.ones(num_nodes),
            jnp.zeros(num_nodes)
        )
    )
    # If not first move, set node_mask_d to zero wherever node_mask_s is 1
    # to avoid same node selection for s and d
    state = state.replace(
        node_mask_d=jax.lax.cond(
            jnp.equal(remaining_actions, total_actions),
            lambda x: x,
            lambda x: jnp.where(
                state.node_mask_s == 1,
                jnp.zeros(num_nodes),
                x
            ),
            state.node_mask_d,
        )
    )

    def mask_previous_selections(i, val):
        # Disallow previously allocated nodes
        update_slice = lambda j, x: jax.lax.dynamic_update_slice_in_dim(x, jnp.array([0.]), j, axis=0)
        val = jax.lax.cond(
            i % 2 == 0,
            lambda x: update_slice(x[0][i], x[1]),  # i is node request index
            lambda x: update_slice(x[0][i+1], x[1]),  # i is slot request index (so add 1 to get next node)
            (state.action_history, val),
        )
        return val

    state = state.replace(
        node_mask_d=jax.lax.fori_loop(
            remaining_actions*2,
            state.action_history.shape[0]-1,
            mask_previous_selections,
            state.node_mask_d
        )
    )
    # If requested node index is new then disallow previously allocated nodes
    # If not new, then must match previously allocated node for that index
    state = state.replace(
        node_mask_d=jax.lax.cond(
            jnp.squeeze(prev_selected_node_d) >= 0,
            lambda x: jnp.where(
                node_indices == prev_selected_node_d,
                x[1],
                x[0],
            ),
            lambda x: x[2],
            (jnp.zeros(num_nodes), jnp.ones(num_nodes), state.node_mask_d),
        )
    )
    return state

mask_slots(state, params, request)

Returns binary mask of valid actions. 1 for valid action, 0 for invalid action.

  1. Check request for source and destination nodes
  2. For each path:
    • Get current slots on path (with padding on end to avoid out of bounds)
    • Get mask for required slots on path
    • Multiply through current slots with required slots mask to check if slots available on path
    • Remove padding from mask
    • Return path mask
  3. Update total mask with path mask
  4. If aggregate_slots > 1, aggregate slot mask to reduce action space

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required
request Array

Request array in format [source_node, data-rate, destination_node]

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
@partial(jax.jit, static_argnums=(1,))
def mask_slots(state: EnvState, params: EnvParams, request: chex.Array) -> EnvState:
    """Returns binary mask of valid actions. 1 for valid action, 0 for invalid action.

    1. Check request for source and destination nodes
    2. For each path:
        - Get current slots on path (with padding on end to avoid out of bounds)
        - Get mask for required slots on path
        - Multiply through current slots with required slots mask to check if slots available on path
        - Remove padding from mask
        - Return path mask
    3. Update total mask with path mask
    4. If aggregate_slots > 1, aggregate slot mask to reduce action space

    Args:
        state: Environment state
        params: Environment parameters
        request: Request array in format [source_node, data-rate, destination_node]

    Returns:
        state: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(request)
    init_mask = jnp.zeros((params.link_resources * params.k_paths))

    def mask_path(i, mask):
        # Get slots for path
        slots = get_path_slots(state.link_slot_array, params, nodes_sd, i)
        # Add padding to slots at end
        slots = jnp.concatenate((slots, jnp.ones(params.max_slots)))
        # Convert bandwidth to slots for each path
        spectral_efficiency = get_paths_se(params, nodes_sd)[i] if params.consider_modulation_format else 1
        requested_slots = required_slots(requested_datarate, spectral_efficiency, params.slot_size, guardband=params.guardband)
        # Get mask used to check if request will fit slots
        request_mask = get_request_mask(requested_slots[0], params)

        def check_slots_available(j, val):
            # Multiply through by request mask to check if slots available
            slot_sum = jnp.sum(request_mask * jax.lax.dynamic_slice(val, (j,), (params.max_slots,))) <= 0
            slot_sum = slot_sum.reshape((1,)).astype(jnp.float32)
            return jax.lax.dynamic_update_slice(val, slot_sum, (j,))

        # Mask out slots that are not valid
        path_mask = jax.lax.fori_loop(
            0,
            int(params.link_resources+1),  # No need to check last requested_slots-1 slots
            check_slots_available,
            slots,
        )
        # Cut off padding
        path_mask = jax.lax.dynamic_slice(path_mask, (0,), (params.link_resources,))
        # Update total mask with path mask
        mask = jax.lax.dynamic_update_slice(mask, path_mask, (i * params.link_resources,))
        return mask

    # Loop over each path
    link_slot_mask = jax.lax.fori_loop(0, params.k_paths, mask_path, init_mask)
    if params.aggregate_slots > 1:
        # Full link slot mask is used in process_path_action to get the correct slot from the aggregated slot action
        state = state.replace(full_link_slot_mask=link_slot_mask)
        link_slot_mask, _ = aggregate_slots(link_slot_mask.reshape(params.k_paths, -1), params)
        link_slot_mask = link_slot_mask.reshape(-1)
    state = state.replace(link_slot_mask=link_slot_mask)
    return state

mask_slots_rmsa_gn_model(state, params, request)

For use in RSAGNModelEnv. 1. For each path: 1.1 Get path slots 1.2 Get launch power

Parameters:

Name Type Description Default
state RSAGNModelEnvState

Environment state

 required
params RSAGNModelEnvParams

Environment parameters

 required
request Array

Request array in format [source_node, data-rate, destination_node]

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
@partial(jax.jit, static_argnums=(1,))
def mask_slots_rmsa_gn_model(state: RSAGNModelEnvState, params: RSAGNModelEnvParams, request: chex.Array) -> EnvState:
    """For use in RSAGNModelEnv.
    1. For each path:
        1.1 Get path slots
        1.2 Get launch power


    Args:
        state: Environment state
        params: Environment parameters
        request: Request array in format [source_node, data-rate, destination_node]

    Returns:
        state: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(request)
    init_mask = jnp.zeros((params.link_resources * params.k_paths))

    def mask_path(i, mask):
        path = get_paths(params, nodes_sd)[i]
        # Get slots for path
        slots = get_path_slots(state.link_slot_array, params, nodes_sd, i)
        # Add padding to slots at end
        # 0 means slot is free, 1 is occupied
        slots = jnp.concatenate((slots, jnp.ones(params.max_slots)))
        launch_power = get_launch_power(state, i, state.launch_power_array[i], params)
        lightpath_index = get_lightpath_index(params, nodes_sd, i)

        # This function checks through each available modulation format, checks the first and last available slots,
        # calculates the SNR, checks it meets the requirements, and returns the resulting mask
        def check_modulation_format(mod_format_index, init_path_mask):
            se = params.modulations_array.val[mod_format_index][1]
            req_slots = required_slots(requested_datarate, se, params.slot_size, guardband=params.guardband)[0]
            bandwidth_per_subchannel = params.slot_size
            req_snr = params.modulations_array.val[mod_format_index][2] + params.snr_margin
            # Get mask used to check if request will fit slots
            request_mask = get_request_mask(req_slots, params)

            def check_slots_available(j, val):
                # Multiply through by request mask to check if slots available
                slot_sum = jnp.sum(request_mask * jax.lax.dynamic_slice(val, (j,), (params.max_slots,))) <= 0
                slot_sum = slot_sum.reshape((1,)).astype(jnp.float32)
                return jax.lax.dynamic_update_slice(val, slot_sum, (j,))

            # Mask out slots that are not valid
            slot_mask = jax.lax.fori_loop(
                0,
                int(params.link_resources + 1),  # No need to check last requested_slots-1 slots
                check_slots_available,
                slots,
            )
            # Cut off padding
            slot_mask = jax.lax.dynamic_slice(slot_mask, (0,), (params.link_resources,))
            # Check first and last available slots for suitability
            ff_path_mask = jnp.concatenate((slot_mask, jnp.ones((1,))), axis=0)
            lf_path_mask = jnp.concatenate((jnp.ones((1,)), slot_mask), axis=0)
            first_available_slot_index = jnp.argmax(ff_path_mask)
            last_available_slot_index = params.link_resources - jnp.argmax(jnp.flip(lf_path_mask)) - 1
            # Assign "req_slots" subchannels (each with bandwidth = slot width) for the first and last possible slots
            ff_temp_state = state.replace(
                channel_centre_bw_array=vmap_set_path_links(state.channel_centre_bw_array, path, first_available_slot_index, req_slots, bandwidth_per_subchannel),
                channel_power_array=vmap_set_path_links(state.channel_power_array, path, first_available_slot_index, req_slots, launch_power),
                path_index_array=vmap_set_path_links(state.path_index_array, path, first_available_slot_index, req_slots, lightpath_index),
                modulation_format_index_array=vmap_set_path_links(state.modulation_format_index_array, path, first_available_slot_index, req_slots, mod_format_index),
            )
            lf_temp_state = state.replace(
                channel_centre_bw_array=vmap_set_path_links(state.channel_centre_bw_array, path, last_available_slot_index, req_slots, bandwidth_per_subchannel),
                channel_power_array=vmap_set_path_links(state.channel_power_array, path, last_available_slot_index, req_slots, launch_power),
                path_index_array=vmap_set_path_links(state.path_index_array, path, last_available_slot_index, req_slots, lightpath_index),
                modulation_format_index_array=vmap_set_path_links(state.modulation_format_index_array, path, last_available_slot_index, req_slots, mod_format_index),
            )
            ff_temp_state = ff_temp_state.replace(link_snr_array=get_snr_link_array(ff_temp_state, params))
            lf_temp_state = lf_temp_state.replace(link_snr_array=get_snr_link_array(lf_temp_state, params))
            # Take the minimum value of SNR from all the subchannels
            ff_snr_value = get_minimum_snr_of_channels_on_path(
                ff_temp_state, path, first_available_slot_index, req_slots, params
            )
            lf_snr_value = get_minimum_snr_of_channels_on_path(
                lf_temp_state, path, last_available_slot_index, req_slots, params
            )
            # Check that other paths SNR is still sufficient (True if failure)
            ff_snr_check = 1 - check_action_rmsa_gn_model(ff_temp_state, None, params)
            lf_snr_check = 1 - check_action_rmsa_gn_model(lf_temp_state, None, params)
            ff_check = (ff_snr_value >= req_snr) * ff_snr_check
            lf_check = (lf_snr_value >= req_snr) * lf_snr_check

            slot_indices = jnp.arange(params.link_resources)
            mod_format_mask = jnp.where(slot_indices == first_available_slot_index, ff_check, False)
            mod_format_mask = jnp.where(slot_indices == last_available_slot_index, lf_check, mod_format_mask)
            path_mask = jnp.where(mod_format_mask, mod_format_index, init_path_mask).astype(jnp.float32)
            # jax.debug.print("ff_snr_check {}", ff_snr_check, ordered=True)
            # jax.debug.print("lf_snr_check {}", lf_snr_check, ordered=True)
            # jax.debug.print("ff_snr_value {}", ff_snr_value, ordered=True)
            # jax.debug.print("lf_snr_value {}", lf_snr_value, ordered=True)
            # jax.debug.print("first_available_slot_index {}", first_available_slot_index, ordered=True)
            # jax.debug.print("last_available_slot_index {}", last_available_slot_index, ordered=True)
            # jax.debug.print("req_snr {}", req_snr, ordered=True)
            # jax.debug.print("mod_format_mask {}", mod_format_mask, ordered=True)
            # jax.debug.print("path_mask {}", path_mask, ordered=True)
            return path_mask

        path_mask = jax.lax.fori_loop(0, params.modulations_array.val.shape[0], check_modulation_format, jnp.full((params.link_resources,), -1.))

        # Update total mask with path mask
        mask = jax.lax.dynamic_update_slice(mask, path_mask, (i * params.link_resources,))
        return mask

    # Loop over each path
    mod_format_mask = jax.lax.fori_loop(0, params.k_paths, mask_path, init_mask)
    link_slot_mask = jnp.where(mod_format_mask >= 0, 1.0, 0.0)
    if params.aggregate_slots > 1:
        # Full link slot mask is used in process_path_action to get the correct slot from the aggregated slot action
        state = state.replace(full_link_slot_mask=link_slot_mask)
        link_slot_mask, _ = aggregate_slots(link_slot_mask.reshape(params.k_paths, -1), params)
        link_slot_mask = link_slot_mask.reshape(-1)
    state = state.replace(
        link_slot_mask=link_slot_mask,
        mod_format_mask=mod_format_mask,
    )
    return state

mask_slots_rwalr(state, params, request)

For use in RWALightpathReuseEnv. Each lightpath has a maximum capacity defined in path_capacity_array. This is updated when a lightpath is assigned. If remaining path capacity is less than current request, corresponding link-slots are masked out. If link-slot is in use by another lightpath for a different source and destination node (even if not full) it is masked out. Step 1: - Mask out slots that are not valid based on path capacity (check link_capacity_array) Step 2: - Mask out slots that are not valid based on lightpath reuse (check path_index_array)

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required
request Array

Request array in format [source_node, data-rate, destination_node]

 required

Returns:

Name Type Description
state EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
@partial(jax.jit, static_argnums=(1,))
def mask_slots_rwalr(state: EnvState, params: EnvParams, request: chex.Array) -> EnvState:
    """For use in RWALightpathReuseEnv.
    Each lightpath has a maximum capacity defined in path_capacity_array. This is updated when a lightpath is assigned.
    If remaining path capacity is less than current request, corresponding link-slots are masked out.
    If link-slot is in use by another lightpath for a different source and destination node (even if not full) it is masked out.
    Step 1:
    - Mask out slots that are not valid based on path capacity (check link_capacity_array)
    Step 2:
    - Mask out slots that are not valid based on lightpath reuse (check path_index_array)

    Args:
        state: Environment state
        params: Environment parameters
        request: Request array in format [source_node, data-rate, destination_node]

    Returns:
        state: Updated environment state
    """
    nodes_sd, requested_datarate = read_rsa_request(request)
    init_mask = jnp.zeros((params.link_resources * params.k_paths))
    source, dest = nodes_sd
    path_start_index = get_path_indices(source, dest, params.k_paths, params.num_nodes, directed=params.directed_graph).astype(jnp.int32)
    #jax.debug.print("path_start_index {}", path_start_index, ordered=True)
    #jax.debug.print("link_capacity_array {}", state.link_capacity_array, ordered=True)

    def mask_path(i, mask):
        # Step 1 - mask capacity
        capacity_mask = jnp.where(state.link_capacity_array < requested_datarate, 1., 0.)
        #jax.debug.print("capacity_mask {}", capacity_mask, ordered=True)
        capacity_slots = get_path_slots(capacity_mask, params, nodes_sd, i)
        #jax.debug.print("capacity_slots {}", capacity_slots, ordered=True)
        # Step 2 - mask lightpath reuse
        lightpath_index = path_start_index + i
        #jax.debug.print("lightpath_index {}", lightpath_index, ordered=True)
        lightpath_mask = jnp.where(state.path_index_array == lightpath_index, 0., 1.)  # Allow current lightpath
        #jax.debug.print("lightpath_mask {}", lightpath_mask, ordered=True)
        lightpath_mask = jnp.where(state.path_index_array == -1, 0., lightpath_mask)  # Allow empty slots
        #jax.debug.print("lightpath_mask {}", lightpath_mask, ordered=True)
        lightpath_slots = get_path_slots(lightpath_mask, params, nodes_sd, i)
        #jax.debug.print("lightpath_slots {}", lightpath_slots, ordered=True)
        # Step 3 combine masks
        path_mask = jnp.max(jnp.stack((capacity_slots, lightpath_slots)), axis=0)
        # Swap zeros for ones
        path_mask = jnp.where(path_mask == 0, 1., 0.)
        #jax.debug.print("path_mask {}", path_mask, ordered=True)
        mask = jax.lax.dynamic_update_slice(mask, path_mask, (i * params.link_resources,))
        return mask

    # Loop over each path
    link_slot_mask = jax.lax.fori_loop(0, params.k_paths, mask_path, init_mask)
    if params.aggregate_slots > 1:
        # Full link slot mask is used in process_path_action to get the correct slot from the aggregated slot action
        state = state.replace(full_link_slot_mask=link_slot_mask)
        link_slot_mask, _ = aggregate_slots(link_slot_mask.reshape(params.k_paths, -1), params)
        link_slot_mask = link_slot_mask.reshape(-1)
    state = state.replace(link_slot_mask=link_slot_mask)
    return state

most_used(state, params, unique_lightpaths, relative)

Get the amount of utilised bandwidth on each lightpath. If RWA-LR environment, the utilisation of a slot is defined by either the count of unique active lightpahts on the slot (if unique_lightpaths is True) or the count of active lightpaths on the slot (if unique_lightpaths is False). If RSA-type environment, utilisation is the count of active lightpaths on that slot.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required
unique_lightpaths bool

Whether to consider unique lightpaths

 required
relative bool

Whether to return relative utilisation

 required

Returns:

Type Description
Array

chex.Array: Most used slots (array length = link_resources)
Source code in xlron/heuristics/heuristics.py 
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
@partial(jax.jit, static_argnums=(1, 2, 3))
def most_used(state: EnvState, params: EnvParams, unique_lightpaths, relative) -> chex.Array:
    """Get the amount of utilised bandwidth on each lightpath.
    If RWA-LR environment, the utilisation of a slot is defined by either the count of unique active lightpahts on the
    slot (if unique_lightpaths is True) or the count of active lightpaths on the slot (if unique_lightpaths is False).
    If RSA-type environment, utilisation is the count of active lightpaths on that slot.

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
        unique_lightpaths (bool): Whether to consider unique lightpaths
        relative (bool): Whether to return relative utilisation

    Returns:
        chex.Array: Most used slots (array length = link_resources)
    """
    if params.__class__.__name__ != "RWALightpathReuseEnvParams":
        most_used_slots = jnp.count_nonzero(state.link_slot_array, axis=0) + 1
    elif params.__class__.__name__ == "RWALightpathReuseEnvParams" and not unique_lightpaths:
        # Get initial path capacity
        initial_path_capacity = init_path_capacity_array(
            params.link_length_array.val, params.path_link_array.val, scale_factor=1.0
        )
        initial_path_capacity = jnp.squeeze(jax.vmap(lambda x: initial_path_capacity[x])(state.path_index_array))
        utilisation = jnp.where(initial_path_capacity - state.link_capacity_array < 0, 0,
                                initial_path_capacity - state.link_capacity_array)
        if relative:
            utilisation = utilisation / initial_path_capacity
        # Get most used slots by summing the utilisation along the slots
        most_used_slots = jnp.sum(utilisation, axis=0) + 1
    else:
        most_used_slots = jnp.count_nonzero(state.path_index_array + 1, axis=0) + 1
    return most_used_slots

mu_ksp(state, params, unique_lightpaths, relative)

Use the most-used available slot on any path. The most-used slot is that which has the most unique lightpaths (if unique_lightpaths=True) or active lightpaths. Method: Go through action mask and find the usage of available slots, choose available slot that is most utilised.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required
params EnvParams

Environment parameters

 required
unique_lightpaths bool

Whether to consider unique lightpaths

 required
relative bool

Whether to return relative utilisation

 required

Returns:

Type Description
Array

chex.Array: Action
Source code in xlron/heuristics/heuristics.py 
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
@partial(jax.jit, static_argnums=(1, 2, 3))
def mu_ksp(state: EnvState, params: EnvParams, unique_lightpaths: bool, relative: bool) -> chex.Array:
    """Use the most-used available slot on any path.
    The most-used slot is that which has the most unique lightpaths (if unique_lightpaths=True) or active lightpaths.
    Method: Go through action mask and find the usage of available slots, choose available slot that is most utilised.

    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
        unique_lightpaths (bool): Whether to consider unique lightpaths
        relative (bool): Whether to return relative utilisation

    Returns:
        chex.Array: Action
    """
    mask = get_action_mask(state, params)
    # Get most used slots by summing the link_slot_array along the links
    most_used_slots = most_used(state, params, unique_lightpaths, relative)
    # Get usage of available slots
    most_used_mask = most_used_slots * mask
    # Chosen slot is the most used globally
    action = jnp.argmax(most_used_mask)
    return action

normalise_traffic_matrix(traffic_matrix)

Normalise traffic matrix to sum to 1

Source code in xlron/environments/env_funcs.py 
580
581
582
583
def normalise_traffic_matrix(traffic_matrix):
    """Normalise traffic matrix to sum to 1"""
    traffic_matrix /= jnp.sum(traffic_matrix)
    return traffic_matrix

pad_array(array, fill_value)

Pad a ragged multidimensional array to rectangular shape. Used for training on multiple topologies. Source: https://codereview.stackexchange.com/questions/222623/pad-a-ragged-multidimensional-array-to-rectangular-shape

Parameters:

Name Type Description Default
array

array to pad

 required
fill_value

value to fill with

 required

Returns:

Name Type Description
result

padded array
Source code in xlron/environments/env_funcs.py 
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
def pad_array(array, fill_value):
    """
    Pad a ragged multidimensional array to rectangular shape.
    Used for training on multiple topologies.
    Source: https://codereview.stackexchange.com/questions/222623/pad-a-ragged-multidimensional-array-to-rectangular-shape

    Args:
        array: array to pad
        fill_value: value to fill with

    Returns:
        result: padded array
    """

    def get_dimensions(array, level=0):
        yield level, len(array)
        try:
            for row in array:
                yield from get_dimensions(row, level + 1)
        except TypeError: #not an iterable
            pass

    def get_max_shape(array):
        dimensions = defaultdict(int)
        for level, length in get_dimensions(array):
            dimensions[level] = max(dimensions[level], length)
        return [value for _, value in sorted(dimensions.items())]

    def iterate_nested_array(array, index=()):
        try:
            for idx, row in enumerate(array):
                yield from iterate_nested_array(row, (*index, idx))
        except TypeError: # final level
            yield (*index, slice(len(array))), array

    dimensions = get_max_shape(array)
    result = np.full(dimensions, fill_value)
    for index, value in iterate_nested_array(array):
        result[index] = value
    return result

path_action_only(topology_pattern, action_counter, remaining_actions)

This is to check if node has already been assigned, therefore just need to assign slots (n=0)

Parameters:

Name Type Description Default
topology_pattern Array

Topology pattern

 required
action_counter Array

Action counter

 required
remaining_actions Scalar

Remaining actions

 required

Returns:

Name Type Description
bool

True if only path action, False if node action
Source code in xlron/environments/env_funcs.py 
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
def path_action_only(topology_pattern: chex.Array, action_counter: chex.Array, remaining_actions: chex.Scalar):
    """This is to check if node has already been assigned, therefore just need to assign slots (n=0)

    Args:
        topology_pattern: Topology pattern
        action_counter: Action counter
        remaining_actions: Remaining actions

    Returns:
        bool: True if only path action, False if node action
    """
    # Get topology segment to be assigned e.g. [2,1,4]
    topology_segment = jax.lax.dynamic_slice(topology_pattern, ((remaining_actions-1)*2, ), (3, ))
    topology_indices = jnp.arange(topology_pattern.shape[0])
    # Check if the latest node in the segment is found in "prev_assigned_topology"
    new_node_to_be_assigned = topology_segment[0]
    prev_assigned_topology = jnp.where(topology_indices > (action_counter[-1]-1)*2, topology_pattern, 0)
    nodes_already_assigned_check = jnp.any(jnp.sum(jnp.where(prev_assigned_topology == new_node_to_be_assigned, 1, 0)) > 0)
    return nodes_already_assigned_check

poisson(key, lam, shape=(), dtype=dtypes.float_)

Sample Exponential random values with given shape and float dtype.

The values are distributed according to the probability density function:

.. math:: f(x) = \lambda e^{-\lambda x}

on the domain :math:0 \le x < \infty.

Args: key: a PRNG key used as the random key. lam: a positive float32 or float64 Tensor indicating the rate parameter shape: optional, a tuple of nonnegative integers representing the result shape. Default (). dtype: optional, a float dtype for the returned values (default float64 if jax_enable_x64 is true, otherwise float32).

Returns: A random array with the specified shape and dtype.

Source code in xlron/environments/env_funcs.py 
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
@partial(jax.jit, static_argnums=(1, 2, 3))
def poisson(key: Union[Array, prng.PRNGKeyArray],
            lam: ArrayLike,
            shape: Shape = (),
            dtype: DTypeLike = dtypes.float_) -> Array:
    r"""Sample Exponential random values with given shape and float dtype.

    The values are distributed according to the probability density function:

    .. math::
     f(x) = \lambda e^{-\lambda x}

    on the domain :math:`0 \le x < \infty`.

    Args:
    key: a PRNG key used as the random key.
    lam: a positive float32 or float64 `Tensor` indicating the rate parameter
    shape: optional, a tuple of nonnegative integers representing the result
      shape. Default ().
    dtype: optional, a float dtype for the returned values (default float64 if
      jax_enable_x64 is true, otherwise float32).

    Returns:
    A random array with the specified shape and dtype.
    """
    key, _ = jax._src.random._check_prng_key(key)
    if not dtypes.issubdtype(dtype, np.floating):
        raise ValueError(f"dtype argument to `exponential` must be a float "
                       f"dtype, got {dtype}")
    dtype = dtypes.canonicalize_dtype(dtype)
    shape = core.canonicalize_shape(shape)
    return _poisson(key, lam, shape, dtype)

process_path_action(state, params, path_action)

Process path action to get path index and initial slot index.

Parameters:

Name Type Description Default
state State

current state

 required
params Params

environment parameters

 required
path_action int

path action

 required

Returns:

Name Type Description
int (Array, Array)

path index
int (Array, Array)

initial slot index
Source code in xlron/environments/env_funcs.py 
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
@partial(jax.jit, static_argnums=(1,))
def process_path_action(state: EnvState, params: EnvParams, path_action: chex.Array) -> (chex.Array, chex.Array):
    """Process path action to get path index and initial slot index.

    Args:
        state (State): current state
        params (Params): environment parameters
        path_action (int): path action

    Returns:
        int: path index
        int: initial slot index
    """
    num_slot_actions = math.ceil(params.link_resources/params.aggregate_slots)
    path_index = jnp.floor(path_action / num_slot_actions).astype(jnp.int32).reshape(1)
    initial_aggregated_slot_index = jnp.mod(path_action, num_slot_actions).reshape(1)
    initial_slot_index = initial_aggregated_slot_index*params.aggregate_slots
    if params.aggregate_slots > 1:
        # Get the path mask do a dynamic slice and get the index of first unoccupied slot in the slice
        path_mask = jax.lax.dynamic_slice(state.full_link_slot_mask, path_index*params.link_resources, (params.link_resources,))
        path_mask_slice = jax.lax.dynamic_slice(path_mask, initial_slot_index, (params.aggregate_slots,))
        # Use argmax to get index of first 1 in slice of mask
        initial_slot_index = initial_slot_index + jnp.argmax(path_mask_slice)
    return path_index[0], initial_slot_index[0]

read_rsa_request(request_array)

Read RSA request from request array. Return source-destination nodes and bandwidth request.

Parameters:

Name Type Description Default
request_array Array

request array

 required

Returns:

Type Description
Tuple[Array, Array]

Tuple[chex.Array, chex.Array]: source-destination nodes and bandwidth request
Source code in xlron/environments/env_funcs.py 
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
def read_rsa_request(request_array: chex.Array) -> Tuple[chex.Array, chex.Array]:
    """Read RSA request from request array. Return source-destination nodes and bandwidth request.

    Args:
        request_array: request array

    Returns:
        Tuple[chex.Array, chex.Array]: source-destination nodes and bandwidth request
    """
    node_s = jax.lax.dynamic_slice(request_array, (0,), (1,))
    requested_datarate = jax.lax.dynamic_slice(request_array, (1,), (1,))
    node_d = jax.lax.dynamic_slice(request_array, (2,), (1,))
    nodes_sd = jnp.concatenate((node_s, node_d))
    return nodes_sd, requested_datarate

remove_expired_node_requests(state)

Check for values in node_departure_array that are less than the current time but greater than zero (negative time indicates the request is not yet finalised). If found, set to infinity in node_departure_array, set to zero in node_resource_array, and increase node_capacity_array by expired resources on each node.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
def remove_expired_node_requests(state: EnvState) -> EnvState:
    """Check for values in node_departure_array that are less than the current time but greater than zero
    (negative time indicates the request is not yet finalised).
    If found, set to infinity in node_departure_array, set to zero in node_resource_array, and increase
    node_capacity_array by expired resources on each node.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.node_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 < state.node_departure_array, mask, 0)
    expired_resources = jnp.sum(jnp.where(mask == 1, state.node_resource_array, 0), axis=1)
    state = state.replace(
        node_capacity_array=state.node_capacity_array + expired_resources,
        node_resource_array=jnp.where(mask == 1, 0, state.node_resource_array),
        node_departure_array=jnp.where(mask == 1, jnp.inf, state.node_departure_array)
    )
    return state

remove_expired_services_rmsa_gn_model(state)

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
def remove_expired_services_rmsa_gn_model(state: EnvState) -> EnvState:
    """

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.link_slot_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 <= state.link_slot_departure_array, mask, 0)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, 0, state.link_slot_array),
        link_slot_departure_array=jnp.where(mask == 1, 0, state.link_slot_departure_array),
        link_snr_array=jnp.where(mask == 1, 0., state.link_snr_array),
        path_index_array=jnp.where(mask == 1, -1, state.path_index_array),
        channel_centre_bw_array=jnp.where(mask == 1, 0, state.channel_centre_bw_array),
        channel_power_array=jnp.where(mask == 1, 0, state.channel_power_array),
        modulation_format_index_array=jnp.where(mask == 1, -1, state.modulation_format_index_array),
        path_index_array_prev=jnp.where(mask == 1, -1, state.path_index_array_prev),
        channel_centre_bw_array_prev=jnp.where(mask == 1, 0, state.channel_centre_bw_array_prev),
        channel_power_array_prev=jnp.where(mask == 1, 0, state.channel_power_array_prev),
        modulation_format_index_array_prev=jnp.where(mask == 1, -1, state.modulation_format_index_array_prev),
    )
    return state

remove_expired_services_rsa(state)

Check for values in link_slot_departure_array that are less than the current time but greater than zero (negative time indicates the request is not yet finalised). If found, set to zero in link_slot_array and link_slot_departure_array.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
def remove_expired_services_rsa(state: EnvState) -> EnvState:
    """Check for values in link_slot_departure_array that are less than the current time but greater than zero
    (negative time indicates the request is not yet finalised).
    If found, set to zero in link_slot_array and link_slot_departure_array.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.link_slot_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 <= state.link_slot_departure_array, mask, 0)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, 0, state.link_slot_array),
        link_slot_departure_array=jnp.where(mask == 1, 0, state.link_slot_departure_array)
    )
    return state

remove_expired_services_rsa_gn_model(state)

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
def remove_expired_services_rsa_gn_model(state: EnvState) -> EnvState:
    """

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.link_slot_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 <= state.link_slot_departure_array, mask, 0)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, 0, state.link_slot_array),
        link_slot_departure_array=jnp.where(mask == 1, 0, state.link_slot_departure_array),
        link_snr_array=jnp.where(mask == 1, 0., state.link_snr_array),
        path_index_array=jnp.where(mask == 1, -1, state.path_index_array),
        channel_centre_bw_array=jnp.where(mask == 1, 0, state.channel_centre_bw_array),
        channel_power_array=jnp.where(mask == 1, 0, state.channel_power_array),
        path_index_array_prev=jnp.where(mask == 1, -1, state.path_index_array_prev),
        channel_centre_bw_array_prev=jnp.where(mask == 1, 0, state.channel_centre_bw_array_prev),
        channel_power_array_prev=jnp.where(mask == 1, 0, state.channel_power_array_prev),
    )
    mask = jnp.where(state.active_lightpaths_array_departure < jnp.squeeze(state.current_time), 1, 0)
    # The active_lightpaths_array is set to -1 when the lightpath is not active
    # The active_lightpaths_array_departure is set to 0 when the lightpath is not active
    # (active_lightpaths_array is used to calculate the total throughput)
    mask = jnp.where(0 <= state.active_lightpaths_array_departure, mask, 0)
    state = state.replace(
        active_lightpaths_array=jnp.where(mask == 1, -1, state.active_lightpaths_array),
        active_lightpaths_array_departure=jnp.where(mask == 1, 0, state.active_lightpaths_array_departure),
    )
    return state

remove_expired_services_rwalr(state)

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
def remove_expired_services_rwalr(state: EnvState) -> EnvState:
    """

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    mask = jnp.where(state.link_slot_departure_array < jnp.squeeze(state.current_time), 1, 0)
    mask = jnp.where(0 <= state.link_slot_departure_array, mask, 0)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, 0, state.link_slot_array),
        link_slot_departure_array=jnp.where(mask == 1, 0, state.link_slot_departure_array),
        path_index_array=jnp.where(mask == 1, 0, state.path_index_array),
    )
    return state

required_slots(bitrate, se, channel_width, guardband=1)

Calculate required slots for a given bitrate and spectral efficiency.

Parameters:

Name Type Description Default
bit_rate float

Bit rate in Gbps

 required
se float

Spectral efficiency in bps/Hz

 required
channel_width float

Channel width in GHz

 required
guardband int

Guard band. Defaults to 1.

 1

Returns:

Name Type Description
int int

Required slots
Source code in xlron/environments/env_funcs.py 
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
@partial(jax.jit, static_argnums=(2, 3))
def required_slots(bitrate: float, se: int, channel_width: float, guardband: int = 1) -> int:
    """Calculate required slots for a given bitrate and spectral efficiency.

    Args:
        bit_rate (float): Bit rate in Gbps
        se (float): Spectral efficiency in bps/Hz
        channel_width (float): Channel width in GHz
        guardband (int, optional): Guard band. Defaults to 1.

    Returns:
        int: Required slots
    """
    # If bitrate is zero, then required slots should be zero
    return jnp.int32((jnp.ceil(bitrate/(se*channel_width))+guardband) * (1 - (bitrate == 0)))

set_c_l_band_gap(link_slot_array, params, val)

Set C+L band gap in link slot array Args: link_slot_array (chex.Array): Link slot array params (RSAGNModelEnvParams): Environment parameters val (int): Value to set Returns: chex.Array: Link slot array with C+L band gap

Source code in xlron/environments/env_funcs.py 
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
@partial(jax.jit, static_argnums=(1, 2))
def set_c_l_band_gap(link_slot_array: chex.Array, params: RSAGNModelEnvParams, val: int) -> chex.Array:
    """Set C+L band gap in link slot array
    Args:
        link_slot_array (chex.Array): Link slot array
        params (RSAGNModelEnvParams): Environment parameters
        val (int): Value to set
    Returns:
        chex.Array: Link slot array with C+L band gap
    """
    # Set C+L band gap
    gap = jnp.full((params.num_links, params.gap_width), val)
    link_slot_array = jax.lax.dynamic_update_slice(link_slot_array, gap, (0, params.gap_start))
    return link_slot_array

undo_action_rmsa_gn_model(state, params)

Undo action for RMSA GN model Args: state (EnvState): Environment state action (chex.Array): Action array params (EnvParams): Environment parameters Returns: EnvState: Updated environment state

Source code in xlron/environments/env_funcs.py 
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
@partial(jax.jit, static_argnums=(1,))
def undo_action_rmsa_gn_model(state: RSAGNModelEnvState, params: RSAGNModelEnvParams) -> EnvState:
    """Undo action for RMSA GN model
    Args:
        state (EnvState): Environment state
        action (chex.Array): Action array
        params (EnvParams): Environment parameters
    Returns:
        EnvState: Updated environment state
    """
    state = undo_action_rsa(state, params)  # Undo link_slot_array and link_slot_departure_array
    state = state.replace(
        link_slot_array=set_c_l_band_gap(state.link_slot_array, params, -1.),  # Set C+L band gap
        channel_centre_bw_array=state.channel_centre_bw_array_prev,
        path_index_array=state.path_index_array_prev,
        channel_power_array=state.channel_power_array_prev,
        modulation_format_index_array=state.modulation_format_index_array_prev,
    )
    return state

undo_action_rsa(state, params)

If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation. Partial resource allocation is indicated by negative time in link slot departure array. Check for values in link_slot_departure_array that are less than zero. If found, increase link_slot_array by +1 and link_slot_departure_array by current_time + holding_time of current request.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
@partial(jax.jit, donate_argnums=(0,))
def undo_action_rsa(state: EnvState, params: Optional[EnvParams]) -> EnvState:
    """If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation.
    Partial resource allocation is indicated by negative time in link slot departure array.
    Check for values in link_slot_departure_array that are less than zero.
    If found, increase link_slot_array by +1 and link_slot_departure_array by current_time + holding_time of current request.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    # If departure array is negative, then undo the action
    mask = jnp.where(state.link_slot_departure_array < 0, 1, 0)
    # If link slot array is < -1, then undo the action
    # (departure might be positive because existing service had holding time after current)
    # e.g. (time_in_array = t1 - t2) where t2 < t1 and t2 = current_time + holding_time
    # but link_slot_array = -2 due to double allocation, so undo the action
    mask = jnp.where(state.link_slot_array < -1, 1, mask)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, state.link_slot_array+1, state.link_slot_array),
        link_slot_departure_array=jnp.where(
            mask == 1,
            state.link_slot_departure_array + state.current_time + state.holding_time,
            state.link_slot_departure_array),
        total_bitrate=state.total_bitrate + read_rsa_request(state.request_array)[1][0]
    )
    return state

undo_action_rsa_gn_model(state, params)

Undo action for RSA GN model Args: state (EnvState): Environment state action (chex.Array): Action array params (EnvParams): Environment parameters Returns: EnvState: Updated environment state

Source code in xlron/environments/env_funcs.py 
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
@partial(jax.jit, static_argnums=(1,))
def undo_action_rsa_gn_model(state: RSAGNModelEnvState, params: RSAGNModelEnvParams) -> EnvState:
    """Undo action for RSA GN model
    Args:
        state (EnvState): Environment state
        action (chex.Array): Action array
        params (EnvParams): Environment parameters
    Returns:
        EnvState: Updated environment state
    """
    state = undo_action_rsa(state, params)  # Undo link_slot_array and link_slot_departure_array
    state = state.replace(
        link_slot_array=set_c_l_band_gap(state.link_slot_array, params, -1.),  # Set C+L band gap
        channel_centre_bw_array=state.channel_centre_bw_array_prev,
        path_index_array=state.path_index_array_prev,
        channel_power_array=state.channel_power_array_prev,
    )
    # If departure array is negative, then undo the action
    mask = jnp.where(state.active_lightpaths_array_departure < 0, 1, 0)
    state = state.replace(
        active_lightpaths_array=jnp.where(mask == 1, -1, state.active_lightpaths_array),
        active_lightpaths_array_departure=jnp.where(
            mask == 1,
            state.active_lightpaths_array_departure + state.current_time + state.holding_time,
            state.active_lightpaths_array_departure),
    )
    return state

undo_action_rwalr(state, params)

If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation. Partial resource allocation is indicated by negative time in link slot departure array. Check for values in link_slot_departure_array that are less than zero. If found, increase link_slot_array by +1 and link_slot_departure_array by current_time + holding_time of current request.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
@partial(jax.jit, donate_argnums=(0,))
def undo_action_rwalr(state: EnvState, params: Optional[EnvParams]) -> EnvState:
    """If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation.
    Partial resource allocation is indicated by negative time in link slot departure array.
    Check for values in link_slot_departure_array that are less than zero.
    If found, increase link_slot_array by +1 and link_slot_departure_array by current_time + holding_time of current request.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    # If departure array is negative, then undo the action
    mask = jnp.where(state.link_slot_departure_array < 0, 1, 0)
    # If link slot array is < -1, then undo the action
    # (departure might be positive because existing service had holding time after current)
    # e.g. (time_in_array = t1 - t2) where t2 < t1 and t2 = current_time + holding_time
    # but link_slot_array = -2 due to double allocation, so undo the action
    mask = jnp.where(state.link_slot_array < -1, 1, mask)
    state = state.replace(
        link_slot_array=jnp.where(mask == 1, state.link_slot_array + 1, state.link_slot_array),
        link_slot_departure_array=jnp.where(
            mask == 1,
            state.link_slot_departure_array + state.current_time + state.holding_time,
            state.link_slot_departure_array),
        total_bitrate=state.total_bitrate + read_rsa_request(state.request_array)[1][0]
    )
    return state

undo_node_action(state)

If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation. Partial resource allocation is indicated by negative time in node departure array. Check for values in node_departure_array that are less than zero. If found, set to infinity in node_departure_array, set to zero in node_resource_array, and increase node_capacity_array by expired resources on each node.

Parameters:

Name Type Description Default
state EnvState

Environment state

 required

Returns:

Type Description
EnvState

Updated environment state
Source code in xlron/environments/env_funcs.py 
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
@partial(jax.jit, donate_argnums=(0,))
def undo_node_action(state: EnvState) -> EnvState:
    """If the request is unsuccessful i.e. checks fail, then remove the partial (unfinalised) resource allocation.
    Partial resource allocation is indicated by negative time in node departure array.
    Check for values in node_departure_array that are less than zero.
    If found, set to infinity in node_departure_array, set to zero in node_resource_array, and increase
    node_capacity_array by expired resources on each node.

    Args:
        state: Environment state

    Returns:
        Updated environment state
    """
    # TODO - Check that node resource clash doesn't happen (so time is always negative after implementation)
    #  and undoing always succeeds with negative time
    mask = jnp.where(state.node_departure_array < 0, 1, 0)
    resources = jnp.sum(jnp.where(mask == 1, state.node_resource_array, 0), axis=1)
    state = state.replace(
        node_capacity_array=state.node_capacity_array + resources,
        node_resource_array=jnp.where(mask == 1, 0, state.node_resource_array),
        node_departure_array=jnp.where(mask == 1, jnp.inf, state.node_departure_array),
    )
    return state

update_action_history(action_history, action_counter, action)

Update action history by adding action to first available index starting from the end.

Parameters:

Name Type Description Default
action_history Array

Action history

 required
action_counter Array

Action counter

 required
action Array

Action to add to history

 required

Returns:

Type Description
Array

Updated action_history
Source code in xlron/environments/env_funcs.py 
850
851
852
853
854
855
856
857
858
859
860
861
def update_action_history(action_history: chex.Array, action_counter: chex.Array, action: chex.Array) -> chex.Array:
    """Update action history by adding action to first available index starting from the end.

    Args:
        action_history: Action history
        action_counter: Action counter
        action: Action to add to history

    Returns:
        Updated action_history
    """
    return jax.lax.dynamic_update_slice(action_history, jnp.flip(action, axis=0).astype(jnp.int32), ((action_counter[-1]-1)*2,))

update_active_lightpaths_array(state, path_index)

Update active lightpaths array with new path index. Find the first index of the array with value -1 and replace with path index. Args: state (RSAGNModelEnvState): Environment state path_index (int): Path index to add to active lightpaths array Returns: jnp.array: Updated active lightpaths array

Source code in xlron/environments/env_funcs.py 
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
def update_active_lightpaths_array(state: RSAGNModelEnvState, path_index: int) -> chex.Array:
    """Update active lightpaths array with new path index.
    Find the first index of the array with value -1 and replace with path index.
    Args:
        state (RSAGNModelEnvState): Environment state
        path_index (int): Path index to add to active lightpaths array
    Returns:
        jnp.array: Updated active lightpaths array
    """
    first_empty_index = jnp.argmin(state.active_lightpaths_array)
    return jax.lax.dynamic_update_slice(state.active_lightpaths_array, jnp.array([path_index]), (first_empty_index,))

update_active_lightpaths_array_departure(state, time)

Update active lightpaths array with new path index. Find the first index of the array with value -1 and replace with path index. Args: state (RSAGNModelEnvState): Environment state time (float): Departure time Returns: jnp.array: Updated active lightpaths array

Source code in xlron/environments/env_funcs.py 
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
def update_active_lightpaths_array_departure(state: RSAGNModelEnvState, time: float) -> chex.Array:
    """Update active lightpaths array with new path index.
    Find the first index of the array with value -1 and replace with path index.
    Args:
        state (RSAGNModelEnvState): Environment state
        time (float): Departure time
    Returns:
        jnp.array: Updated active lightpaths array
    """
    first_empty_index = jnp.argmin(state.active_lightpaths_array)
    return jax.lax.dynamic_update_slice(state.active_lightpaths_array_departure, time, (first_empty_index,))

update_graph_tuple(state, params)

Update graph tuple for use with Jraph GNNs. Edge and node features are updated from link_slot_array and node_capacity_array respectively. Global features are updated as request_array. Args: state (EnvState): Environment state params (EnvParams): Environment parameters Returns: state (EnvState): Environment state with updated graph tuple

Source code in xlron/environments/env_funcs.py 
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
def update_graph_tuple(state: EnvState, params: EnvParams):
    """Update graph tuple for use with Jraph GNNs.
    Edge and node features are updated from link_slot_array and node_capacity_array respectively.
    Global features are updated as request_array.
    Args:
        state (EnvState): Environment state
        params (EnvParams): Environment parameters
    Returns:
        state (EnvState): Environment state with updated graph tuple
    """
    # Get source and dest from request array
    source_dest, _ = read_rsa_request(state.request_array)
    source, dest = source_dest[0], source_dest[2]
    # Current request as global feature
    globals = jnp.reshape(state.request_array, (1, -1))
    # One-hot encode source and destination
    source_dest_features = jnp.zeros((params.num_nodes, 2))
    source_dest_features = source_dest_features.at[source.astype(jnp.int32), 0].set(1)
    source_dest_features = source_dest_features.at[dest.astype(jnp.int32), 1].set(-1)
    spectral_features = state.graph.nodes[..., :3]

    if params.__class__.__name__ in ["RSAGNModelEnvParams", "RMSAGNModelEnvParams"]:
        # Normalize by max parameters (converted to linear units)
        max_power = isrs_gn_model.from_dbm(params.max_power)
        normalized_power = jnp.round(state.channel_power_array / max_power, 3)
        max_snr = isrs_gn_model.from_db(params.max_snr)
        normalized_snr = jnp.round(state.link_snr_array / max_snr, 3)
        edge_features = jnp.stack([normalized_snr, normalized_power], axis=-1)
        node_features = jnp.concatenate([spectral_features, source_dest_features], axis=-1)
    elif params.__class__.__name__ == "VONEEnvParams":
        edge_features = state.link_slot_array
        node_features = getattr(state, "node_capacity_array", jnp.zeros(params.num_nodes))
        node_features = node_features.reshape(-1, 1)
        node_features = jnp.concatenate([node_features, spectral_features, source_dest_features], axis=-1)
    else:
        edge_features = state.link_slot_array
        node_features = jnp.concatenate([spectral_features, source_dest_features], axis=-1)

    if params.disable_node_features:
        node_features = jnp.zeros((1,))

    edge_features = edge_features if params.directed_graph else jnp.repeat(edge_features, 2, axis=0)
    graph = state.graph._replace(nodes=node_features, edges=edge_features, globals=globals)
    state = state.replace(graph=graph)
    return state

update_node_array(node_indices, array, node, request)

Used to udated selected_nodes array with new requested resources on each node, for use in

Source code in xlron/environments/env_funcs.py 
973
974
975
def update_node_array(node_indices, array, node, request):
    """Used to udated selected_nodes array with new requested resources on each node, for use in """
    return jnp.where(node_indices == node, array-request, array)

Set relevant slots along links in path to val.

Parameters:

Name Type Description Default
link_slot_array Array

Link slot array

 required
path Array

Path (row from path-link array that indicates links used by path)

 required
initial_slot int

Initial slot

 required
num_slots int

Number of slots

 required
value int

Value to set on link slot array

 required

Returns:

Type Description
Array

Updated link slot array
Source code in xlron/environments/env_funcs.py 
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
@jax.jit
def vmap_set_path_links(link_slot_array: chex.Array, path: chex.Array, initial_slot: int, num_slots: int, value: int) -> chex.Array:
    """Set relevant slots along links in path to val.

    Args:
        link_slot_array: Link slot array
        path: Path (row from path-link array that indicates links used by path)
        initial_slot: Initial slot
        num_slots: Number of slots
        value: Value to set on link slot array

    Returns:
        Updated link slot array
    """
    return jax.vmap(set_path, in_axes=(0, 0, None, None, None))(link_slot_array, path, initial_slot, num_slots, value)

vmap_update_node_departure(node_departure_array, selected_nodes, value)

Called when implementing node action. Sets request departure time ("value") in place of first "inf" i.e. unoccupied index on node departure array for selected nodes.

Parameters:

Name Type Description Default
node_departure_array Array

(N x R) Node departure array

 required
selected_nodes Array

(N x 1) Selected nodes (non-zero value on selected node indices)

 required
value int

Value to set on node departure array

 required

Returns:

Type Description
Array

Updated node departure array
Source code in xlron/environments/env_funcs.py 
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
@jax.jit
def vmap_update_node_departure(node_departure_array: chex.Array, selected_nodes: chex.Array, value: int) -> chex.Array:
    """Called when implementing node action.
    Sets request departure time ("value") in place of first "inf" i.e. unoccupied index on node departure array for selected nodes.

    Args:
        node_departure_array: (N x R) Node departure array
        selected_nodes: (N x 1) Selected nodes (non-zero value on selected node indices)
        value: Value to set on node departure array

    Returns:
        Updated node departure array
    """
    first_inf_indices = jnp.argmax(node_departure_array, axis=1)
    return jax.vmap(update_selected_node_departure, in_axes=(0, 0, 0, None))(node_departure_array, selected_nodes, first_inf_indices, value)

vmap_update_node_resources(node_resource_array, selected_nodes)

Called when implementing node action. Sets requested node resources on selected nodes in place of first "zero" i.e. unoccupied index on node resource array for selected nodes.

Parameters:

Name Type Description Default
node_resource_array

(N x R) Node resource array

 required
selected_nodes

(N x 1) Requested resources on selected nodes

 required

Returns:

Type Description

Updated node resource array
Source code in xlron/environments/env_funcs.py 
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
@jax.jit
def vmap_update_node_resources(node_resource_array, selected_nodes):
    """Called when implementing node action.
    Sets requested node resources on selected nodes in place of first "zero" i.e.
    unoccupied index on node resource array for selected nodes.

    Args:
        node_resource_array: (N x R) Node resource array
        selected_nodes: (N x 1) Requested resources on selected nodes

    Returns:
        Updated node resource array
    """
    first_zero_indices = jnp.argmin(node_resource_array, axis=1)
    return jax.vmap(update_selected_node_resources, in_axes=(0, 0, 0))(node_resource_array, selected_nodes, first_zero_indices)

Update relevant slots along links in path to current_val - val.

Parameters:

Name Type Description Default
link_slot_array Array

Link slot array

 required
path Array

Path (row from path-link array that indicates links used by path)

 required
initial_slot int

Initial slot

 required
num_slots int

Number of slots

 required
value int

Value to subtract from link slot array

 required

Returns:

Type Description
Array

Updated link slot array
Source code in xlron/environments/env_funcs.py 
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
@jax.jit
def vmap_update_path_links(link_slot_array: chex.Array, path: chex.Array, initial_slot: int, num_slots: int, value: int) -> chex.Array:
    """Update relevant slots along links in path to current_val - val.

    Args:
        link_slot_array: Link slot array
        path: Path (row from path-link array that indicates links used by path)
        initial_slot: Initial slot
        num_slots: Number of slots
        value: Value to subtract from link slot array

    Returns:
        Updated link slot array
    """
    return jax.vmap(update_path, in_axes=(0, 0, None, None, None))(link_slot_array, path, initial_slot, num_slots, value)