pyneat.genome module¶
This module defines the genome encoding used by NEAT.
-
class
pyneat.genome.ConnectionGene(key, node_in, node_out, weight, expressed)[source]¶ Bases:
objectDefines a connection gene used in the genome encoding.
-
expressed¶ True if the connection is expressed (enabled) in the phenotype, False otherwise.
Type: bool
-
copy()[source]¶ Create a copy of the connection gene.
Returns: A copy of itself. Return type: ConnectionGene
-
-
class
pyneat.genome.Genome(key, config, innovation_store)[source]¶ Bases:
objectDefines a genome used to encode a neural network.
-
config¶ The genome configuration settings.
Type: GenomeConfig
-
inputs¶ The node keys of input nodes.
Type: :list:`int`
-
outputs¶ The node keys of output nodes.
Type: :list:`int`
-
biases¶ The node keys of bias nodes.
Type: :list:`int`
-
innovation_store¶ The global innovation store used for tracking new structural mutations.
Type: InnovationStore
-
add_bias_node(num)[source]¶ Add a new bias node.
Parameters: num (int) – A number that can uniquely identify bias nodes in the innovation store.
-
add_connection(node_in, node_out, weight, expressed=True)[source]¶ Add a connection between two nodes.
Parameters: - node_in (int) – The key of the node that leads into the new connection.
- node_out (int) – The key of the node that the the new connection leads into.
- weight (float) – The weight of the connection. Must be a value between [0, 1].
- expressed (bool) – True if the connection should be expressed in the phenotype, False otherwise.
-
add_node(node_in, node_out, node_type)[source]¶ Add a new node positioned between two other nodes. Input and output nodes are positioned between non-existent nodes.
Parameters: Returns: The key of the new node
Return type:
-
configure_crossover(parent1, parent2, average)[source]¶ Performs crossover between two genomes.
If the two genomes have equal fitness then the joint and excess genes are inherited from the smaller genome.
Parameters:
-
configure_new()[source]¶ Configure a new genome based on the given configuration.
The initial inputs and outputs for input and output nodes are specified as negatives so that matching innovation keys are generated for corresponding input and output nodes between genomes. Inputs nodes use odd negative numbers, and output nodes use even negative numbers.
-
copy()[source]¶ Create a copy of the genome.
Note: Copies share the same config and innovation store.
Returns: - A copy of itself, but with the same config and innovation
- store.
Return type: Genome
-
distance(other)[source]¶ Computes the compatibility (genetic) distance between two genomes.
This is used for deciding how to speciate the population. Distance is a function of the number of disjoint and excess genes, as well as the weight/bias differences of matching genes.
Update (13.04.20): Distance is measured using the original compatibility distance measure defined by Stanley & Miikkulainen (2002).
Parameters: other (Genome) – The other genome to compare itself to. Returns: The genetic distance between itself and the other genome. Return type: float
-
mutate()[source]¶ Mutate the genome.
Mutates the genome according to the mutation parameter values specified in the genome configuration.
As per the original implementation of NEAT:
- If any structural mutations are performed, weight and bias mutations will not be performed.
- If an add node mutation is performed, an add connection mutation will not also be performed.
-
mutate_add_connection()[source]¶ Performs an ‘add connection’ structural mutation.
A single connection with a random weight is added between two previously unconnected nodes.
Returns: True if a connection was added, False otherwise. Return type: bool
-
mutate_add_node()[source]¶ Performs an ‘add node’ structural mutation.
An existing connection is split and the new node is placed where the old connection used to be. The old connection is disabled and two new connection genes are added. The new connection leading into the new node receives a weight of 1.0 and the connection leading out of the new node receives the old connection weight.
Connections from bias nodes and non-expressed nodes are not split.
Returns: True is a node was added, False otherwise. Return type: bool
-
mutate_weights()[source]¶ Mutates (perturbs) or replaces each connection weight in the genome.
Each weight is either replaced (with some probability, specified in the genome config) or perturbed.
Replaced weights and perturbations are drawn from a uniform distribution with range [-weight_perturb_power, weight_perturb_power).
-
classmethod
parse_config(param_dict)[source]¶ Takes a dictionary of configuration items, returns an object that will later be passed to the write_config method.
Parameters: param_dict (dict) – A dictionary of configuration parameter values. Returns: The genome configuration. Return type: GenomeConfig
-
size()[source]¶ Returns a measure of genome complexity.
Returns: - A measure of the complexity of the genome given by
- (number of nodes, number of enabled connections)
Return type: tuple
-
classmethod
write_config(filename, config)[source]¶ Takes a file-like object and the configuration object created by parse_config. This method should write the configuration item definitions to the given file.
Parameters: - filename (str) – The name of the file to write the genome configuration to.
- config (GenomeConfig) – The genome configuration to save.
-
-
class
pyneat.genome.GenomeConfig(params)[source]¶ Bases:
objectSets up and holds configuration information for the Genome class.
- Config Parameters:
num_inputs (int): The number of inputs each network should have.
num_outputs (int): The number of outputs each network should have.
num_biases (int): The number of bias nodes the network should have.
initial_conn_prob (float): The initial connection probability of each potential connection between inputs and outputs. 0.0 = no connections, i.e. all inputs are disconnected from the outputs. 1.0 = fully connected, i.e. all inputs are connected to all outputs.
activation_func (str): The name of the activation function to be used by hidden and output nodes. Must be present in the set of possible activation functions.
compatibility_disjoint_coefficient (float): The disjoint and excess coefficient to be used when calculating genome distance.
compatibility_weight_coefficient (float): The weight and bias coefficient to be used when calculation genome distance.
normalise_gene_dist (bool): Whether or not normalise the gene dist (for genetic distance calculations) for large genomes.
feed_forward (bool): False if recurrent connections are allowed, True otherwise.
conn_add_prob (float): The probability of adding a new connection when performing mutations.
node_add_prob (float): The probability of adding a new node when performing mutations.
weight_mutate_prob (float): The probability of mutating the connection weights of a genome when performing mutations.
weight_replace_prob (float): The probability of replacing, instead of perturbing, a connection weight when performing weight mutations.
weight_init_power (float): Sets the range of possible values for weight replacements and new weight initialisations.
weight_perturb_power (float): Sets the range of possible values for weight perturbations.
weight_min_value (float): Sets the minimum allowed value for connection weights.
weight_max_value (float): Sets the maximum allowed value for connection weights.
gene_disable_prob (float): The probability of disabling a gene in the child that is disabled in either of the parents when performing crossover.