def given_another_genome():
node1 = NodeGene(NodeGeneType.INPUT, 1, 1)
node2 = NodeGene(NodeGeneType.INPUT, 2, 2)
node3 = NodeGene(NodeGeneType.INPUT, 3, 3)
node4 = NodeGene(NodeGeneType.OUTPUT, 4, 4)
node5 = NodeGene(NodeGeneType.HIDDEN, 5, 5)
node6 = NodeGene(NodeGeneType.HIDDEN, 6, 6)
nodes = {
node1.get_id(): node1,
node2.get_id(): node2,
node3.get_id(): node3,
node4.get_id(): node4,
node5.get_id(): node5,
node6.get_id(): node6
}
conn1 = ConnectionGene(node1.get_id(),
node4.get_id(),
0.8,
innovation_number=1)
conn2 = ConnectionGene(node2.get_id(), node4.get_id(), -0.4, False, 2)
conn3 = ConnectionGene(node3.get_id(),
node4.get_id(),
0.5,
innovation_number=3)
conn4 = ConnectionGene(node2.get_id(),
node5.get_id(),
0.2,
innovation_number=4)
conn5 = ConnectionGene(node5.get_id(),
node6.get_id(),
0.4,
innovation_number=5)
conn6 = ConnectionGene(node3.get_id(),
node5.get_id(),
0.6,
innovation_number=6)
conn7 = ConnectionGene(node1.get_id(),
node6.get_id(),
0.6,
innovation_number=7)
conn8 = ConnectionGene(node6.get_id(),
node4.get_id(),
0.6,
innovation_number=8)
connections = {
conn1.get_innovation_number(): conn1,
conn2.get_innovation_number(): conn2,
conn3.get_innovation_number(): conn3,
conn4.get_innovation_number(): conn4,
conn5.get_innovation_number(): conn5,
conn6.get_innovation_number(): conn6,
conn7.get_innovation_number(): conn7,
conn8.get_innovation_number(): conn8
}
genome = Genome(connections, nodes, InnovationGenerator())
return genome
def test_distance():
genome_less_fitted = given_a_genome()
genome_more_ftted = given_another_genome()
distance = Genome.compute_distance(genome_less_fitted, genome_more_ftted,
1, 1, 1)
print(distance)
def given_a_genome():
node1 = NodeGene(NodeGeneType.INPUT, 0, 0)
node2 = NodeGene(NodeGeneType.INPUT, 1, 1)
node3 = NodeGene(NodeGeneType.INPUT, 2, 2)
node4 = NodeGene(NodeGeneType.OUTPUT, 3, 3)
node5 = NodeGene(NodeGeneType.HIDDEN, 4, 4)
nodes = {
node1.get_id(): node1,
node2.get_id(): node2,
node3.get_id(): node3,
node4.get_id(): node4,
node5.get_id(): node5
}
conn1 = ConnectionGene(node1.get_id(),
node4.get_id(),
0.7,
innovation_number=1)
conn2 = ConnectionGene(node2.get_id(), node4.get_id(), -0.5, False, 2)
conn3 = ConnectionGene(node3.get_id(),
node4.get_id(),
0.5,
innovation_number=3)
conn4 = ConnectionGene(node2.get_id(),
node5.get_id(),
0.2,
innovation_number=4)
conn5 = ConnectionGene(node5.get_id(),
node4.get_id(),
0.4,
innovation_number=5)
conn6 = ConnectionGene(node1.get_id(),
node5.get_id(),
0.6,
innovation_number=6)
conn7 = ConnectionGene(node4.get_id(),
node5.get_id(),
0.6,
False,
innovation_number=11)
connections = {
conn1.get_innovation_number(): conn1,
conn2.get_innovation_number(): conn2,
conn3.get_innovation_number(): conn3,
conn4.get_innovation_number(): conn4,
conn5.get_innovation_number(): conn5,
conn6.get_innovation_number(): conn6,
conn7.get_innovation_number(): conn7
}
inno = InnovationGenerator()
inno.counter = 11
genome = Genome(connections, nodes, inno)
return genome
def is_same_species(parent_1_genome, species):
from NEAT.Genome import Genome
"""
Returns True if the genomes is in same species else False
Threshold value = 3.0 Refer paper section 4.1
:param parent_1_genome: The genome of anyone to be added in this species
:param species: Species object
:return: Boolean representing whether genome belong in the provided species
"""
threshold = 3.0
return True if Genome.get_compatibility_distance(
parent_1_genome, species.genomes[0]) < threshold else False
def initialize_population(self):
"""
Initializes population by creating random genomes and adding them to appropriate species
"""
from NEAT.Genome import Genome
import numpy as np
birds = list()
for i in range(self.population_size):
random_genome_connections = Genome.get_random_connection_genes(
input_nodes=2, output_nodes=1, init=True)
genome = Genome(random_genome_connections,
input_nodes=2,
output_nodes=1)
self.add_to_species(genome)
birds.append(
Bird(100,
np.random.randint(20, 500),
genome=genome,
show_bird=True))
return birds
def crossover(parent_1_genome, parent_2_genome):
from NEAT.Genome import Genome
"""
Abstract for crossover between two genomes.
Actual crossover at get_child_connections(parent_1_genome, parent_2_genome)
Mutation rates: Refer paper section 4.1
Node Mutation = 0.03
Connection Mutation = 0.05
Weight Mutation = 0.8
Weight Perturb Rate = 0.9
Weight Reassign Rate = 0.1
Connection Enable/Disable Rate = 0.75
Inter-species Crossover = 0.001
:param parent_1_genome: The genome for 1st parent
:param parent_2_genome: The genome for 2nd parent
:return: Child genome with mutation applied
"""
inter_species_crossover_rate = 0.001
node_mutation_rate = 0.03
connection_mutation_rate = 0.05
connection_enable_disable_rate = 0.75
weight_mutation = 0.8
weight_perturb_rate = 0.9
weight_reassign_rate = 0.1
if Species.is_same_species(parent_1_genome, parent_2_genome):
perform_crossover = True
else:
perform_crossover = True if np.random.uniform(
) < inter_species_crossover_rate else False
if perform_crossover:
child_connections = Species.get_child_connections(
parent_1_genome, parent_2_genome)
child_genome = Genome(child_connections)
if np.random.uniform() < node_mutation_rate:
child_genome.add_node_mutation()
if np.random.uniform() < connection_mutation_rate:
child_genome.add_connection_mutation()
if np.random.uniform() < weight_mutation:
child_genome.perturb_weights(weight_perturb_rate)
child_genome.reassign_weights(weight_reassign_rate)
if np.random.uniform() < connection_enable_disable_rate:
child_genome.enable_disable_gene(
connection_enable_disable_rate)
return child_genome
else:
return max([parent_1_genome, parent_2_genome],
key=lambda genome: genome.fitness)
def test_crossover():
genome1 = given_a_genome()
genome2 = given_another_genome()
new_genome = Genome.crossover(genome2, genome1)
print(new_genome)