Wirings

class ncps.wirings.Wiring(units)

Bases: object

property num_layers

get_neurons_of_layer(layer_id)

is_built()

build(input_dim)

erev_initializer(shape=None, dtype=None)

sensory_erev_initializer(shape=None, dtype=None)

set_input_dim(input_dim)

set_output_dim(output_dim)

get_type_of_neuron(neuron_id)

add_synapse(src, dest, polarity)

add_sensory_synapse(src, dest, polarity)

get_config()

classmethod from_config(config)

get_graph(include_sensory_neurons=True)

Returns a networkx.DiGraph object of the wiring diagram :param include_sensory_neurons: Whether to include the sensory neurons as nodes in the graph

property synapse_count

Counts the number of synapses between internal neurons of the model

property sensory_synapse_count

Counts the number of synapses from the inputs (sensory neurons) to the internal neurons of the model

draw_graph(layout='shell', neuron_colors=None, synapse_colors=None, draw_labels=False)

Draws a matplotlib graph of the wiring structure Examples:

>>> import matplotlib.pyplot as plt
>>> plt.figure(figsize=(6, 4))
>>> legend_handles = wiring.draw_graph(draw_labels=True)
>>> plt.legend(handles=legend_handles, loc="upper center", bbox_to_anchor=(1, 1))
>>> plt.tight_layout()
>>> plt.show()

Parameters:

• layout –

• neuron_colors –

• synapse_colors –

• draw_labels –

Returns:

class ncps.wirings.AutoNCP(units, output_size, sparsity_level=0.5, seed=22222)

Bases: NCP

Instantiate an NCP wiring with only needing to specify the number of units and the number of outputs

Parameters:

• units – The total number of neurons

• output_size – The number of motor neurons (=output size). This value must be less than units-2 (typically good choices are 0.3 times the total number of units)

• sparsity_level – A hyperparameter between 0.0 (very dense) and 0.9 (very sparse) NCP.

• seed – Random seed for generating the wiring

class ncps.wirings.NCP(inter_neurons, command_neurons, motor_neurons, sensory_fanout, inter_fanout, recurrent_command_synapses, motor_fanin, seed=22222)

Bases: Wiring

Creates a Neural Circuit Policies wiring. The total number of neurons (= state size of the RNN) is given by the sum of inter, command, and motor neurons. For an easier way to generate a NCP wiring see the AutoNCP wiring class.

Parameters:

• inter_neurons – The number of inter neurons (layer 2)

• command_neurons – The number of command neurons (layer 3)

• motor_neurons – The number of motor neurons (layer 4 = number of outputs)

• sensory_fanout – The average number of outgoing synapses from the sensory to the inter neurons

• inter_fanout – The average number of outgoing synapses from the inter to the command neurons

• recurrent_command_synapses – The average number of recurrent connections in the command neuron layer

• motor_fanin – The average number of incoming synapses of the motor neurons from the command neurons

• seed – The random seed used to generate the wiring

property num_layers

get_neurons_of_layer(layer_id)

get_type_of_neuron(neuron_id)

build(input_shape)

class ncps.wirings.FullyConnected(units, output_dim=None, erev_init_seed=1111, self_connections=True)

Bases: Wiring

build(input_shape)

class ncps.wirings.Random(units, output_dim=None, sparsity_level=0.0, random_seed=1111)

Bases: Wiring

build(input_shape)