def test_sort_into_species(functions_object, genome_four_nodes,
genome_six_nodes):
config = get_config()
i_four_1 = individual.Individual(genome_four_nodes, config['kBias_node'])
i_four_2 = individual.Individual(genome_four_nodes, config['kBias_node'])
i_four_3 = individual.Individual(genome_four_nodes, config['kBias_node'])
i_six_1 = individual.Individual(genome_six_nodes, config['kBias_node'])
i_six_2 = individual.Individual(genome_six_nodes, config['kBias_node'])
i_four_2.assigned_specie = "Not Assigned"
i_six_1.assigned_specie = "Not Assigned"
s1 = species.Species(0, i_four_1, config=get_config())
individuals = [i_four_1, i_four_2, i_four_3, i_six_1, i_six_2]
all_species = {s1.id: s1}
assert i_four_2.assigned_specie == "Not Assigned"
assert i_six_1.assigned_specie == "Not Assigned"
functions_object.sort_into_species(all_species, individuals, 1)
assert len(all_species) == 2
assert all_species[2].generation_created == 1
assert all_species[2].representative is i_six_1
assert len(all_species[1].members) == 4
assert len(all_species[2].members) == 2
assert i_four_2.assigned_specie == 1
assert i_six_1.assigned_specie == 2
def species_one_member(genome_basic):
config = get_config()
i1 = individual.Individual(genome_basic, config['kBias_node'])
i1.fitness = 1
s1 = species.Species(0, i1, config=get_config())
s1.members = [i1]
return s1
def test_mate_inverted_fitness(genome_four_nodes, genome_five_nodes):
config = get_config()
i1 = Individual(genome_four_nodes, config['kBias_node'])
i2 = Individual(genome_five_nodes, config['kBias_node'])
i1.fitness = 5
i2.fitness = 10
species_object = Species(0, i1, config=get_config())
offspring = species_object.mate(i1, i2)
assert offspring.genome.get_genome_size() == i2.genome.get_genome_size()
for innov_num in range(1, 12):
assert innov_num in offspring.genome.get_all_gene_ids()
def genome_basic():
config = get_config()
c1 = genome.ConnectionGene(1, 2, 0.1, innov_num=3)
n1 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=1)
n2 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='output',
innov_num=2)
connections = {
c1.innov_num: c1,
}
nodes = {
n1.innov_num: n1,
n2.innov_num: n2,
}
g1 = genome.Genome(config=config)
g1.node_genes = nodes
g1.connection_genes = connections
return g1
def species_five_members(genome_basic):
config = get_config()
i1 = individual.Individual(genome_basic, config['kBias_node'])
i2 = individual.Individual(genome_basic, config['kBias_node'])
i3 = individual.Individual(genome_basic, config['kBias_node'])
i4 = individual.Individual(genome_basic, config['kBias_node'])
i5 = individual.Individual(genome_basic, config['kBias_node'])
i1.fitness = 10
i2.fitness = 15
i3.fitness = 40
i4.fitness = 80
i5.fitness = 5
s1 = species.Species(0, i1, config=get_config())
s1.members = [i1, i2, i3, i4, i5]
return s1
def genome_simple_shuffle():
config = get_config()
c1 = genome.ConnectionGene(1, 3, 0.5, innov_num=5)
c2 = genome.ConnectionGene(2, 3, 0.5, innov_num=6)
c3 = genome.ConnectionGene(1, 4, 0.1, innov_num=7)
c4 = genome.ConnectionGene(4, 3, 0.8, innov_num=8)
c5 = genome.ConnectionGene(9, 3, 0.25, innov_num=10)
c6 = genome.ConnectionGene(4, 9, 0.333, innov_num=11)
c7 = genome.ConnectionGene(2, 12, 0.22, innov_num=13)
c8 = genome.ConnectionGene(12, 4, 0.22, innov_num=14)
c9 = genome.ConnectionGene(1, 15, 0.1, innov_num=16)
c10 = genome.ConnectionGene(15, 3, 0.1, innov_num=17)
n1 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=1)
n2 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=2)
n3 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='output',
innov_num=3)
n4 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'], innov_num=4)
n5 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'], innov_num=9)
n6 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'], innov_num=12)
n7 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'], innov_num=15)
connections = {
c1.innov_num: c1,
c2.innov_num: c2,
c3.innov_num: c3,
c4.innov_num: c4,
c5.innov_num: c5,
c6.innov_num: c6,
c7.innov_num: c7,
c8.innov_num: c8,
c9.innov_num: c9,
c10.innov_num: c10
}
nodes = {
n1.innov_num: n1,
n2.innov_num: n2,
n3.innov_num: n3,
n4.innov_num: n4,
n5.innov_num: n5,
n6.innov_num: n6,
n7.innov_num: n7
}
g1 = genome.Genome(config=config)
g1.node_genes = nodes
g1.connection_genes = connections
return g1
def test_add_member(species_six_members, genome_basic):
config = get_config()
new_member = Individual(genome_basic, config['kBias_node'])
assert new_member.assigned_specie is None
assert len(species_six_members.members) == 6
species_six_members.add_member(new_member)
assert new_member.assigned_specie == species_six_members.id
assert len(species_six_members.members) == 7
def genome_six_nodes():
config = get_config()
c1 = genome.ConnectionGene(1, 3, 0.6, innov_num=5)
c2 = genome.ConnectionGene(2, 3, 0.75, innov_num=6)
c3 = genome.ConnectionGene(1, 4, 0.6, innov_num=7)
c4 = genome.ConnectionGene(4, 3, -0.1, innov_num=8)
c5 = genome.ConnectionGene(1, 9, 0.1, innov_num=10)
c6 = genome.ConnectionGene(9, 3, 0.1, innov_num=11)
c7 = genome.ConnectionGene(1, 12, 0.3, innov_num=13)
c8 = genome.ConnectionGene(12, 4, 0.1, innov_num=14)
n1 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=1)
n2 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=2)
n3 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='output',
innov_num=3)
n4 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'], innov_num=4)
n5 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'], innov_num=9)
n6 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'], innov_num=12)
connections = {
c1.innov_num: c1,
c2.innov_num: c2,
c3.innov_num: c3,
c4.innov_num: c4,
c5.innov_num: c5,
c6.innov_num: c6,
c7.innov_num: c7,
c8.innov_num: c8
}
nodes = {
n1.innov_num: n1,
n2.innov_num: n2,
n3.innov_num: n3,
n4.innov_num: n4,
n5.innov_num: n5,
n6.innov_num: n6
}
g1 = genome.Genome(config=config)
g1.node_genes = nodes
g1.connection_genes = connections
return g1
def genome_recursive_two_outputs():
config = get_config()
c1 = genome.ConnectionGene(1, 3, 1.0, innov_num=5)
c2 = genome.ConnectionGene(2, 3, 1.0, innov_num=6)
c3 = genome.ConnectionGene(1, 4, -1.0, innov_num=7)
c4 = genome.ConnectionGene(2, 4, -1.0, innov_num=8)
c5 = genome.ConnectionGene(1, 9, 0.0, innov_num=10)
c6 = genome.ConnectionGene(9, 4, 0.5, innov_num=11)
c7 = genome.ConnectionGene(9, 9, 1.0, innov_num=12)
n1 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=1)
n2 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=2)
n3 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='output',
innov_num=3)
n4 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='output',
innov_num=4)
n5 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'], innov_num=9)
connections = {
c1.innov_num: c1,
c2.innov_num: c2,
c3.innov_num: c3,
c4.innov_num: c4,
c5.innov_num: c5,
c6.innov_num: c6,
c7.innov_num: c7
}
nodes = {
n1.innov_num: n1,
n2.innov_num: n2,
n3.innov_num: n3,
n4.innov_num: n4,
n5.innov_num: n5
}
g1 = genome.Genome(config=config)
g1.node_genes = nodes
g1.connection_genes = connections
return g1
def genome_isolated_output_node():
config = get_config()
c1 = genome.ConnectionGene(1, 3, 0.1, innov_num=5)
c2 = genome.ConnectionGene(2, 3, 0.25, innov_num=6)
c3 = genome.ConnectionGene(1, 4, 0.1, innov_num=7)
c4 = genome.ConnectionGene(2, 4, -0.6, innov_num=8)
n1 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=1)
n2 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=2)
n3 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='output',
innov_num=3)
n4 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='output',
innov_num=4)
# Isolate output node 4
c3.enabled = False
c4.enabled = False
connections = {
c1.innov_num: c1,
c2.innov_num: c2,
c3.innov_num: c3,
c4.innov_num: c4
}
nodes = {
n1.innov_num: n1,
n2.innov_num: n2,
n3.innov_num: n3,
n4.innov_num: n4,
}
g1 = genome.Genome(config=config)
g1.node_genes = nodes
g1.connection_genes = connections
return g1
def genome_four_nodes_recursive():
config = get_config()
c1 = genome.ConnectionGene(1, 3, 0.1, innov_num=5)
c2 = genome.ConnectionGene(2, 3, 0.25, innov_num=6)
c3 = genome.ConnectionGene(1, 4, 0.1, innov_num=7)
c4 = genome.ConnectionGene(4, 3, -0.6, innov_num=8)
c5 = genome.ConnectionGene(4, 4, 0.25, innov_num=9) # Recursive connection
n1 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=1)
n2 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='input',
innov_num=2)
n3 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'],
node_type='output',
innov_num=3)
n4 = genome.NodeGene(sigmoid_power=config['kSigmoid_power'], innov_num=4)
connections = {
c1.innov_num: c1,
c2.innov_num: c2,
c3.innov_num: c3,
c4.innov_num: c4,
c5.innov_num: c5
}
nodes = {
n1.innov_num: n1,
n2.innov_num: n2,
n3.innov_num: n3,
n4.innov_num: n4,
}
g1 = genome.Genome(config=config)
g1.node_genes = nodes
g1.connection_genes = connections
return g1
def __init__(self, inputs, outputs, _config=None):
super().__init__()
self.config = get_config() if not _config else _config
self.functions = Functions(config=self.config)
self.persistence = persistence.Persistence(self.config)
self.generation = 0
self.population = self._get_initial_population(inputs, outputs)
self.current_unused_innov = self.population[
0].genome.get_greatest_innov() + 1
self.session_stats = visualization.Visualization()
self.current_champion = None
# TESTING:
self.all_innovs = dict()
first_species = species.Species(self.generation,
self.population[0],
config=self.config)
self.species = {first_species.id: first_species}
for individual in self.population[1:]:
self.species[first_species.id].add_member(individual)
while not game_over:
output_vector = individual.frame(board_state)
strongest_signal = output_vector.index(max(output_vector))
game_over, board_state = tetris_game.do_frame(
strongest_signal)
individual.fitness = tetris_game.score
self.child_conn.send(sub_population)
print("Ending Process")
if __name__ == '__main__':
xor_config = get_config("tetris_config")
session = NEATSession(200, 5, xor_config)
parent_conn, child_conn = Pipe()
parent_conn2, child_conn2 = Pipe()
parent_conn3, child_conn3 = Pipe()
parent_conn4, child_conn4 = Pipe()
p = ParallelTetrisPlayer(child_conn)
p2 = ParallelTetrisPlayer(child_conn2)
p3 = ParallelTetrisPlayer(child_conn3)
p4 = ParallelTetrisPlayer(child_conn4)
p.start()
p2.start()
p3.start()
def functions_object():
f1 = functions.Functions(config=get_config())
return f1
}
options = list(inputs.keys())
random.shuffle(options)
for choice in options:
inputs[choice] = individual.evaluate(inputs[choice])[0]
squared_performance = ((1 - inputs['neither']) + inputs['left'] + inputs['right'] + (1 - inputs['both'])) ** 2
_individual.fitness = squared_performance
if __name__ == '__main__':
target_fitness = 14
xor_config = get_config("XOR_config")
session = NEATSession(2, 1, xor_config)
for generation in range(10000):
population = session.get_individuals()
for individual in population:
evaluate_task(individual)
session.collect_individuals(population)
session.advance_generation()
print(session.current_champion.fitness, "Number of species: {0}".format(len(session.species)))
if session.current_champion.fitness > target_fitness:
print("Success on generation {0}".format(generation))
session.show_stats(session.current_champion)
break
class Persistence(object):
def __init__(self, config):
super().__init__()
self.config = config
self.genome_folder = self._get_genome_directory_path()
def _get_genome_directory_path(self):
#Make the folder if it doesn't exist
final_path = self.config['Project Root']
specified_folder = self.config['kGenome_folder'].split(".")
while specified_folder:
final_path = os.path.join(final_path, specified_folder.pop(0))
return final_path
def save_genomes(self, genomes):
# Save out a list of gnomes to that folder with a unique timestamp
print(self.config['Project Root'])
pass
def load_genomes(self):
# Try to load any genomes from the specified directory
return "testing"
if __name__ == '__main__':
config = get_config()
test_persist = Persistence(config)
test_persist.save_genomes('test')
