def _weights_differences_avg(genome_1: Genome, genome_2: Genome) -> float:
diff = 0
for con_1 in genome_1.connections():
for con_2 in genome_2.connections():
if con_1 == con_2:
diff += np.abs(con_1.weight() - con_2.weight())
return diff
def construct(genome: Genome, file_name):
dot = Digraph()
for node in genome.nodes():
dot.node(str(node.id()), str(node.id()))
for con in genome.connections():
if con.status() == Status.ENABLED:
dot.edge(str(con.input_node().id()), str(con.output_node().id()))
dot.render(f'/Users/max/IdeaProjects/neat/network_pictures/{file_name}',
view=True)
def _excesses_disjoints(genome_1: Genome, genome_2: Genome):
E = 0
D = 0
last_disjoint = _pick_last_disjoint_connection(genome_1, genome_2)
for con in genome_1.connections() + genome_2.connections():
if con not in genome_1.connections() or con not in genome_2.connections():
if con.innovation_number() <= last_disjoint:
D += 1
else:
E += 1
return E, D
def test_init_1(self):
print("\n\ntest_init_1.1")
generation = Generation()
genome = Genome(generation=generation, input_nodes=2, output_nodes=2)
for con in genome.connections():
print(con)
for node in genome.nodes():
print(node)
print('-' * 40)
for node in generation.nodes():
print(node)
print('-' * 60)
viz.construct(genome, 'first')
print("\n\ntest_init_1.2")
genome_1 = Genome(generation, 2, 2)
for con in genome_1.connections():
print(con)
for node in genome_1.nodes():
print(node)
print('-' * 40)
for node in generation.nodes():
print(node)
viz.construct(genome_1, 'second')
print('-' * 20 + 'Mutations')
for mutation in generation.mutations():
print(mutation)
def test_add_connection(self):
print('\n\ntest_add_connection\nBEFORE 0')
generation = Generation()
genome = Genome(generation, 2, 1)
for con in genome.connections():
print(con)
viz.construct(genome, 'before_0')
genome._add_node()
genome._add_connection()
genome._add_connection()
genome._add_connection()
print('\nAFTER 0\n')
for con in genome.connections():
print(con)
viz.construct(genome, 'after_0')
def spawn(self, first=True, input_neurons=0, output_neurons=1):
if first:
for i in range(population_size):
# Spawn new Genome with number of input/output neurons
genome = Genome(self, input_neurons, output_neurons)
self.add_organism(genome)
else:
pass
def test_excesses_and_disjoints(self):
generation = Generation()
genome_1 = Genome(generation, 1, 2)
genome_2 = genome_1.copy_genome()
viz.construct(genome_1, 'First Before')
genome_1._add_node()
genome_1._add_connection()
for con in genome_1.connections():
print(con)
viz.construct(genome_1, 'First')
genome_2._add_node()
viz.construct(genome_2, 'Second')
E, D = _excesses_disjoints(genome_1, genome_2)
print('-' * 20)
for con in genome_2.connections():
print(con)
print(f'Excesses: {E}')
print(f'Disjoints: {D}')
def test_predict(self):
generation = Generation()
genome = Genome(generation=generation, input_nodes=2, output_nodes=2)
nn = NeuralNetwork(genome=genome)
construct(nn.genome(), 'Before')
nn.add_node()
construct(nn.genome(), '1 Node Added')
nn.add_node()
construct(nn.genome(), '2 Nodes Added')
nn.add_connection()
construct(nn.genome(), 'Connection Added')
print("CONNECTIONS")
for con in nn.genome().connections():
print(con)
print("CONNECTIONS")
for pred in nn.predict([2, 3]):
print(pred)
def test_init_2(self):
generation = Generation()
genome = Genome(generation=generation, input_nodes=1, output_nodes=1)
genome._add_node()
new_generation = Generation(_copy=True, prev_generation=generation)
genome_copied = Genome(generation=new_generation,
_copy=True,
genome_to_copy=genome)
for node in generation.nodes():
print(node)
print('-' * 20)
viz.construct(genome, 'first_1')
genome_copied._add_node()
for node in new_generation.nodes():
print(node)
viz.construct(genome_copied, 'first_copied_1')
def test_sigma(self):
generation = Generation()
genome_1 = Genome(generation, 1, 1)
genome_2 = genome_1.copy_genome()
# genome_1._add_node()
genome_1._mutate_weights()
genome_2._mutate_weights()
for con in genome_1.connections():
print(con)
print('-' * 20)
for con in genome_2.connections():
print(con)
print(sigma(genome_1, genome_2))
def test_offspring(self):
generation = Generation()
genome_1 = Genome(generation, 2, 2)
genome_2 = Genome(generation, 2, 2)
# for i in range(2):
# genome_1._add_connection()
# genome_2._add_connection()
# for i in range(1):
# genome_1._add_node()
# genome_2._add_node()
viz.construct(genome_1, 'Genome 1')
viz.construct(genome_2, 'Genome 2')
for con in genome_1.connections():
print(con)
print('-' * 40)
for con in genome_2.connections():
print(con)
genome = produce_offspring(generation, genome_1, genome_2)
print('-' * 40)
for node in genome.nodes():
print(node)
print('\n' + '-' * 40)
for con in genome.connections():
print(con)
viz.construct(genome, 'offspring')
E, D = _excesses_disjoints(genome_1, genome_2)
print('-' * 20)
print(f'Excesses: {E}')
print(f'Disjoints: {D}')
def spawn(self):
for i in range(population_size):
genome = Genome(self, number_input_neurons, number_output_neurons)
nn = NeuralNetwork(genome)
self.add_organism(nn)
def produce_offspring(generation: Generation, genome_1: Genome, genome_2: Genome) -> Genome:
offspring = Genome(generation, genome_1.input_nodes(), genome_1.output_nodes(), connections=False)
for node in genome_1.nodes() + genome_2.nodes():
if node not in offspring.nodes():
offspring.nodes().append(node.copy_without_connections())
for con in genome_1.connections() + genome_2.connections():
if (con in offspring.connections()) and (con.status() == Status.DISABLED):
for i, in_con in enumerate(offspring.connections()):
if in_con == con and in_con.status() == Status.ENABLED:
if _disable_chance():
in_con.disable()
break
elif con not in offspring.connections():
if (con.status() == Status.DISABLED) and (not _disable_chance()):
con.enable()
offspring.connections().append(con)
return offspring
def test_add_node(self):
print('-' * 20 + 'test_add_node\nBEFORE 0')
generation = Generation()
genome = Genome(generation, 2, 1)
for con in genome.connections():
print(con)
for node in genome.nodes():
print(node)
viz.construct(genome, 'before_0')
genome._add_node()
genome._add_node()
genome._add_node()
for node in generation.nodes():
print(node)
print('\nAFTER 0 \n')
for con in genome.connections():
print(con)
print('-' * 40)
for node in genome.nodes():
print(node)
viz.construct(genome, 'after_0')
print('\n\ntest_add_node\nBEFORE 1')
genome_1 = Genome(generation, 1, 1)
viz.construct(genome_1, 'before_1')
for con in genome_1.connections():
print(con)
for node in genome_1.nodes():
print(node)
genome_1._add_node()
genome_1._add_node()
genome_1._add_node()
genome_1._add_node()
print('\nAFTER 0 \n')
for con in genome_1.connections():
print(con)
for node in genome_1.nodes():
print(node)
viz.construct(genome_1, 'after_1')
def produce_offspring(generation, genome_1: Genome,
genome_2: Genome) -> Genome:
# TODO: exclude excess genes of parent 2 if parent 1 is fitter
offspring = Genome(generation,
genome_1.input_nodes(),
genome_1.output_nodes(),
connections=False)
for node in genome_1.nodes() + genome_2.nodes():
if node not in offspring.nodes():
offspring.nodes().append(node.copy_without_connections())
# all common genes are appended randomly from either parent
for con in genome_1.connections():
if con in genome_2.connections():
index = genome_2.connections().index(con)
gene = _pick_random_gene(con, genome_2.connections()[index])
_append_con(gene, offspring)
for con in genome_1.connections() + genome_2.connections():
if (con in offspring.connections()) and (con.status()
== Status.DISABLED):
for in_con in offspring.connections():
if in_con == con and in_con.status() == Status.ENABLED:
if _disable_chance():
in_con.disable()
break
elif con not in offspring.connections():
if (con.status() == Status.DISABLED) and (not _disable_chance()):
con.enable()
_append_con(con, offspring)
return offspring
