def create_recurrent_phenotype(genome):
""" Receives a genome and returns its phenotype (a recurrent neural network). """
# Gather inputs and expressed connections.
input_nodes = [
ng.ID for ng in genome.node_genes.values() if ng.type == 'INPUT'
]
output_nodes = [
ng.ID for ng in genome.node_genes.values() if ng.type == 'OUTPUT'
]
node_inputs = {}
used_nodes = set(input_nodes + output_nodes)
for cg in genome.conn_genes.values():
if not cg.enabled:
continue
used_nodes.add(cg.out_node_id)
used_nodes.add(cg.in_node_id)
if cg.out_node_id not in node_inputs:
node_inputs[cg.out_node_id] = [(cg.in_node_id, cg.weight)]
else:
node_inputs[cg.out_node_id].append((cg.in_node_id, cg.weight))
node_evals = []
for onode, inputs in node_inputs.items():
ng = genome.node_genes[onode]
activation_function = activation_functions.get(ng.activation_type)
node_evals.append(
(onode, activation_function, ng.bias, ng.response, inputs))
return RecurrentNetwork(max(used_nodes), input_nodes, output_nodes,
node_evals)
def create_feed_forward_phenotype(genome):
""" Receives a genome and returns its phenotype (a neural network). """
# Gather inputs and expressed connections.
input_nodes = [
ng.ID for ng in genome.node_genes.values() if ng.type == 'INPUT'
]
output_nodes = [
ng.ID for ng in genome.node_genes.values() if ng.type == 'OUTPUT'
]
connections = [(cg.in_node_id, cg.out_node_id)
for cg in genome.conn_genes.values() if cg.enabled]
layers = find_feed_forward_layers(input_nodes, connections)
node_evals = []
used_nodes = set(input_nodes + output_nodes)
for layer in layers:
for node in layer:
inputs = []
# TODO: This could be more efficient.
for cg in genome.conn_genes.values():
if cg.out_node_id == node and cg.enabled:
inputs.append((cg.in_node_id, cg.weight))
used_nodes.add(cg.in_node_id)
used_nodes.add(node)
ng = genome.node_genes[node]
activation_function = activation_functions.get(ng.activation_type)
node_evals.append(
(node, activation_function, ng.bias, ng.response, inputs))
return FeedForwardNetwork(max(used_nodes), input_nodes, output_nodes,
node_evals)
def create_feed_forward_phenotype(genome):
""" Receives a genome and returns its phenotype (a neural network). """
# Gather inputs and expressed connections.
input_nodes = [ng.ID for ng in genome.node_genes.values() if ng.type == 'INPUT']
output_nodes = [ng.ID for ng in genome.node_genes.values() if ng.type == 'OUTPUT']
connections = [(cg.in_node_id, cg.out_node_id) for cg in genome.conn_genes.values() if cg.enabled]
layers = find_feed_forward_layers(input_nodes, connections)
node_evals = []
used_nodes = set(input_nodes + output_nodes)
for layer in layers:
for node in layer:
inputs = []
# TODO: This could be more efficient.
for cg in genome.conn_genes.values():
if cg.out_node_id == node and cg.enabled:
inputs.append((cg.in_node_id, cg.weight))
used_nodes.add(cg.in_node_id)
used_nodes.add(node)
ng = genome.node_genes[node]
activation_function = activation_functions.get(ng.activation_type)
node_evals.append((node, activation_function, ng.bias, ng.response, inputs))
return FeedForwardNetwork(max(used_nodes), input_nodes, output_nodes, node_evals)
def create_recurrent_phenotype(genome):
""" Receives a genome and returns its phenotype (a recurrent neural network). """
# Gather inputs and expressed connections.
input_nodes = [ng.ID for ng in genome.node_genes.values() if ng.type == 'INPUT']
output_nodes = [ng.ID for ng in genome.node_genes.values() if ng.type == 'OUTPUT']
node_inputs = {}
used_nodes = set(input_nodes + output_nodes)
for cg in genome.conn_genes.values():
if not cg.enabled:
continue
used_nodes.add(cg.out_node_id)
used_nodes.add(cg.in_node_id)
if cg.out_node_id not in node_inputs:
node_inputs[cg.out_node_id] = [(cg.in_node_id, cg.weight)]
else:
node_inputs[cg.out_node_id].append((cg.in_node_id, cg.weight))
node_evals = []
for onode, inputs in node_inputs.items():
ng = genome.node_genes[onode]
activation_function = activation_functions.get(ng.activation_type)
node_evals.append((onode, activation_function, ng.bias, ng.response, inputs))
return RecurrentNetwork(max(used_nodes), input_nodes, output_nodes, node_evals)
def __init__(self, genome, config):
"""
FeedForwardPhenome - A feedforward network
Adapted from: https://github.com/CodeReclaimers/neat-python, accessed May 2016
:param genome: the genome to create the phenome
"""
self.graph, node_lists = self._construct_graph(genome)
self.input_nodes, self.hidden_nodes, self.output_nodes = node_lists
self.links = [(g.src, g.sink) for g in genome.link_genes]
self.node_evals = []
self.config = config
layers = find_feed_forward_layers(self.input_nodes, self.links)
used_nodes = set(self.input_nodes + self.output_nodes)
for layer in layers:
for node in layer:
inputs = []
# TODO: This could be more efficient.
for cg in genome.link_genes:
if cg.sink == node and cg.enabled:
inputs.append((cg.src, cg.weight))
used_nodes.add(cg.src)
used_nodes.add(node)
ng = genome.get_node_by_index(node)
activation_function = activation_functions.get(ng.activation)
self.node_evals.append((node, activation_function, inputs))
self.values = [0.0] * (1 + max(used_nodes))
def __init__(self, neuron_type, ID, bias, response, activation_type):
assert neuron_type in ('INPUT', 'OUTPUT', 'HIDDEN')
self.type = neuron_type
self.ID = self.indexer.next(ID)
self.bias = bias
self.response = response
self.activation = activation_functions.get(activation_type)
self._synapses = []
self.output = 0.0 # for recurrent networks all neurons must have an "initial state"
def __init__(self, neuron_type, ID, bias, response, activation_type):
assert neuron_type in ('INPUT', 'OUTPUT', 'HIDDEN')
self.type = neuron_type
self.ID = self.indexer.next(ID)
self.bias = bias
self.response = response
self.activation = activation_functions.get(activation_type)
self._synapses = []
self.output = 0.0 # for recurrent networks all neurons must have an "initial state"