def test_generation(self):
genome1 = create_genome_structure(
5, 2, modified_sigmoid_activation, NeatConfig(),
InnovationNumberGeneratorSingleCore())
genome2 = create_genome_structure(
5, 2, modified_sigmoid_activation, NeatConfig(),
InnovationNumberGeneratorSingleCore())
agent1 = Agent(1, genome1)
agent2 = Agent(2, genome2)
members = [agent1, agent2]
species = [
Species(1,
genome2, [agent1, agent2],
max_species_fitness=10,
generation_max_species_fitness=3,
adjust_fitness=8)
]
seed = 10
generation = Generation(2, seed, members, species)
self.assertEqual(2, generation.number)
self.assertEqual(species, generation.species_list)
self.assertEqual(members, generation.agents)
self.assertEqual(seed, generation.seed)
def test_mutate_add_connection(self):
config = NeatConfig(connection_initial_min_weight=-3,
connection_initial_max_weight=3,
allow_recurrent_connections=True,
probability_mutate_add_connection=0.4,
mutate_connection_tries=2)
self.genome.connections = []
# Random values for with seed 1
# rnd.uniform(0, 1) = 0.417022004702574 -> No mutation
_, conn = mutate_add_connection(self.genome, self.rnd, self.inn_generator, config)
self.assertEqual(0, len(self.genome.connections))
self.assertIsNone(conn)
# Set higher config value, reset random
config.probability_mutate_add_connection = 1
self.rnd = np.random.RandomState(1)
# Random values for with seed 1
# rnd.uniform(0, 1) = 0.417022004702574 -> Mutate
# rnd.randint(low=0,high=4) = 0
# rnd.randint(low=0,high=2) = 0
# rnd.uniform(low=-3, high=3) = -2.9993137510959307
_, new_con1 = mutate_add_connection(self.genome, self.rnd, self.inn_generator, config)
self.assertEqual(1, len(self.genome.connections))
self.assertEqual(4, new_con1.innovation_number)
self.assertEqual(self.node_input1.innovation_number, new_con1.input_node)
self.assertEqual(self.node_hidden1.innovation_number, new_con1.output_node)
self.assertAlmostEqual(-2.9993137510959307, new_con1.weight, delta=0.0000000001)
# Random values
# rnd.uniform(0, 1) = 0.00011437481734488664 -> Mutate
# rnd.randint(low=0,high=4) = 3
# rnd.randint(low=0,high=2) = 0
# rnd.uniform(low=-3, high=3) = -2.445968431387213
_, new_con2 = mutate_add_connection(self.genome, self.rnd, self.inn_generator, config)
self.assertEqual(2, len(self.genome.connections))
self.assertEqual(5, new_con2.innovation_number)
self.assertEqual(self.node_output1.innovation_number, new_con2.input_node)
self.assertEqual(self.node_hidden1.innovation_number, new_con2.output_node)
self.assertAlmostEqual(-2.445968431387213, new_con2.weight, delta=0.0000000001)
# Brute force - check if all nodes are connected after 1000 tries
for _ in range(1000):
_, _ = mutate_add_connection(self.genome, self.rnd, self.inn_generator, config)
# Maximum connections with recurrent
self.assertEqual(8, len(self.genome.connections))
# Set recurrent to false
config.allow_recurrent = False
genome_feed_forward = Genome(1, self.nodes, connections=[])
for _ in range(1000):
genome_feed_forward, _ = mutate_add_connection(genome_feed_forward, self.rnd, self.inn_generator, config)
# Maximum connections with recurrent=false
self.assertEqual(5, len(genome_feed_forward.connections))
def test_mutate_add_node(self):
config = NeatConfig(probability_mutate_add_node=0.1, bias_initial_min=-3, bias_initial_max=3)
genome, node, con1, con2 = mutate_add_node(self.genome, self.rnd, self.inn_generator, config)
# Should not mutate
self.assertEqual(self.genome, genome)
self.assertIsNone(node)
self.assertIsNone(con1)
self.assertIsNone(con2)
# Set probability high, and reset randomState
config.probability_mutate_add_node = 1
self.rnd = np.random.RandomState(1)
self.node_output1.activation_function = modified_sigmoid_activation
# Random values for with seed 1
# rnd.uniform(0, 1) = 0.417022004702574 -> Mutate
# rnd.randint(low=0,high=4) = 0 -> connection1
# rnd.uniform(0, 1) = -0.9325573593386588 -> Take activation function of out node
# rnd.uniform(-3, 3) = -2.23125331242386 -> Bias for new node
node_size_before = len(self.genome.nodes)
connections_size_before = len(self.connections)
genome, node, con1, con2 = mutate_add_node(self.genome, self.rnd, self.inn_generator, config)
# Check if 1 node and 2 connections were added
self.assertEqual(genome, self.genome)
self.assertEqual(node_size_before + 1, len(genome.nodes))
self.assertEqual(connections_size_before + 2, len(genome.connections))
# Check if old connection was disabled
self.assertFalse(self.connection1.enabled)
# Check the generated node
self.assertEqual(NodeType.HIDDEN, node.node_type)
self.assertEqual(modified_sigmoid_activation, node.activation_function)
self.assertEqual(4, node.innovation_number)
self.assertAlmostEqual(-2.23125331242386, node.bias, delta=0.0000000001)
self.assertEqual(0.5, node.x_position)
# Check 1 connection
self.assertEqual(4, con1.innovation_number)
self.assertEqual(self.connection1.input_node, con1.input_node)
self.assertEqual(node.innovation_number, con1.output_node)
self.assertEqual(1, con1.weight)
self.assertTrue(con1.enabled)
# Check 2 connection
self.assertEqual(5, con2.innovation_number)
self.assertEqual(node.innovation_number, con2.input_node)
self.assertEqual(self.connection1.output_node, con2.output_node)
self.assertEqual(self.connection1.weight, con2.weight)
self.assertTrue(con2.enabled)
def setUp(self) -> None:
self.optimizer_single = NeatOptimizerSingleCore()
self.callback = MockCallback()
self.challenge = MockChallenge()
self.optimizer_single.register_callback(self.callback)
self.config = NeatConfig(population_size=150)
def test_create_offspring_pairs(self):
agent_id_generator = AgentIDGeneratorSingleCore()
config = NeatConfig()
rnd = np.random.RandomState(1)
# First 10 random values
# rnd.randint(3) = 1
# rnd.randint(3) = 0
# rnd.randint(3) = 0
# rnd.randint(3) = 1
# rnd.randint(3) = 1
# rnd.randint(3) = 0
# rnd.randint(3) = 0
# rnd.randint(3) = 1
tuple_list = ss.create_offspring_pairs(self.species1, 4,
agent_id_generator,
self.generation, rnd, config)
# Test tuples
agent_id_generator = AgentIDGeneratorSingleCore()
self.assertEqual(4, len(tuple_list))
self.assertEqual((self.agent2.id, self.agent1.id,
agent_id_generator.get_agent_id()), tuple_list[0])
self.assertEqual((self.agent1.id, self.agent2.id,
agent_id_generator.get_agent_id()), tuple_list[1])
self.assertEqual((self.agent2.id, self.agent1.id,
agent_id_generator.get_agent_id()), tuple_list[2])
self.assertEqual((self.agent1.id, self.agent2.id,
agent_id_generator.get_agent_id()), tuple_list[3])
def test_config(self):
config = NeatConfig(population_size=200,
connection_min_weight=-10,
connection_max_weight=10)
self.assertEqual(200, config.population_size)
self.assertEqual(-10, config.connection_min_weight)
self.assertEqual(10, config.connection_max_weight)
def test_species(self):
genome1 = create_genome_structure(5, 2, modified_sigmoid_activation, NeatConfig(),
InnovationNumberGeneratorSingleCore())
genome2 = create_genome_structure(5, 2, modified_sigmoid_activation, NeatConfig(),
InnovationNumberGeneratorSingleCore())
members = [Agent(1, genome1)]
species = Species(1, genome2, members, max_species_fitness=10, generation_max_species_fitness=3,
adjust_fitness=8)
self.assertEqual(genome2, species.representative)
self.assertEqual(8, species.adjusted_fitness)
self.assertEqual(10, species.max_species_fitness)
self.assertEqual(3, species.generation_max_species_fitness)
self.assertEqual(members, species.members)
self.assertEqual(1, species.id_)
def test_set_new_genome_bias(self):
config = NeatConfig(bias_initial_min=-3, bias_initial_max=3)
modified_genome = set_new_genome_bias(self.genome, self.rnd, config)
# Input nodes should be skipped
self.assertEqual(0, modified_genome.nodes[0].bias)
self.assertEqual(0, modified_genome.nodes[1].bias)
self.assertAlmostEqual(-0.4978679717845562, modified_genome.nodes[2].bias, delta=0.0000001)
self.assertAlmostEqual(1.3219469606529488, modified_genome.nodes[3].bias, delta=0.0000001)
def test_cross_over(self):
node1_1 = Node(1, NodeType.INPUT, 1.1, step_activation, 0)
node1_2 = Node(2, NodeType.INPUT, 1.2, step_activation, 0)
node1_5 = Node(5, NodeType.OUTPUT, 1.5, step_activation, 1)
node1_7 = Node(7, NodeType.HIDDEN, 1.7, step_activation, 0.5)
nodes1 = [node1_1, node1_2, node1_5, node1_7]
node2_1 = Node(1, NodeType.INPUT, 2.1, modified_sigmoid_activation, 0)
node2_2 = Node(2, NodeType.INPUT, 2.2, modified_sigmoid_activation, 0)
node2_4 = Node(4, NodeType.OUTPUT, 2.4, modified_sigmoid_activation, 1)
node2_7 = Node(7, NodeType.HIDDEN, 2.7, modified_sigmoid_activation, 0.5)
node2_8 = Node(8, NodeType.HIDDEN, 2.8, modified_sigmoid_activation, 0.5)
nodes2 = [node2_1, node2_2, node2_4, node2_7, node2_8]
connection1_1 = Connection(innovation_number=1, input_node=1, output_node=2, weight=1.2, enabled=False)
connection1_2 = Connection(innovation_number=2, input_node=1, output_node=2, weight=1.2, enabled=True)
connection1_4 = Connection(innovation_number=4, input_node=1, output_node=2, weight=1.2, enabled=True)
connection1_7 = Connection(innovation_number=7, input_node=1, output_node=2, weight=1.2, enabled=False)
connections1 = [connection1_1, connection1_2, connection1_4, connection1_7]
connection2_1 = Connection(innovation_number=1, input_node=1, output_node=2, weight=1.2, enabled=False)
connection2_2 = Connection(innovation_number=2, input_node=1, output_node=2, weight=1.2, enabled=True)
connection2_3 = Connection(innovation_number=3, input_node=1, output_node=2, weight=1.2, enabled=True)
connection2_5 = Connection(innovation_number=5, input_node=1, output_node=2, weight=1.2, enabled=True)
connection2_7 = Connection(innovation_number=7, input_node=1, output_node=2, weight=1.2, enabled=False)
connections2 = [connection2_1, connection2_2, connection2_3, connection2_5, connection2_7]
more_fit_parent = Genome(1, nodes1, connections1)
less_fit_parent = Genome(2, nodes2, connections2)
config = NeatConfig(probability_enable_gene=0.31)
# Random values:
# 1 rnd.uniform(0, 1) = 0.417022004702574 -> First node (node1_1)
# 1 rnd.uniform(0, 1) = 0.7203244934421581 -> Second node(node2_2)
# 1 rnd.uniform(0, 1) = 0.00011437481734488664 -> First node (node7_1)
# 1 rnd.uniform(0, 1) = 0.30233257263183977 -> Select first connection (connection1_1)
# 1 rnd.uniform(0, 1) = 0.14675589081711304 -> Re-enable connection (connection1_1)
# 1 rnd.uniform(0, 1) = 0.0923385947687978 -> Select first connection (connection1_2)
# 1 rnd.uniform(0, 1) = 0.1862602113776709 -> Select first connection (connection1_7)
# 1 rnd.uniform(0, 1) = 0.34556072704304774 -> Re-enable connection False (connection1_7)
child_nodes, child_connections = cross_over(more_fit_parent, less_fit_parent, self.rnd, config)
# Check nodes
self.assertEqual(4, len(child_nodes))
self.compare_nodes(node1_1, child_nodes[0])
self.compare_nodes(node2_2, child_nodes[1])
self.compare_nodes(node1_5, child_nodes[2])
self.compare_nodes(node1_7, child_nodes[3])
# Check connections
self.assertEqual(4, len(child_connections))
connection1_1.enabled = True
self.compare_connections(connection1_1, child_connections[0])
self.compare_connections(connection1_2, child_connections[1])
self.compare_connections(connection1_4, child_connections[2])
self.compare_connections(connection1_7, child_connections[3])
def evaluate(self, optimizer: NeatOptimizer, seed: int = None, **kwargs):
time = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
# Reporter for fitness values
self.fitness_reporter = FitnessReporter()
self.species_reporter = SpeciesReporter()
self.time_reporter = TimeReporter()
self.check_point_reporter = CheckPointReporter(
"/tmp/breakout/{}/".format(time), "breakout", lambda _: True)
# Register this class as callback
optimizer.register_callback(self)
optimizer.register_reporters(self.time_reporter, self.species_reporter,
self.fitness_reporter,
self.check_point_reporter)
# Config
config = NeatConfig(allow_recurrent_connections=False,
population_size=150,
compatibility_threshold=3,
weight_mutation_type="normal",
weight_mutation_normal_sigma=1.3,
connection_initial_min_weight=-1,
connection_initial_max_weight=1,
connection_min_weight=-15,
connection_max_weight=15,
bias_mutation_type="normal",
bias_mutation_normal_sigma=1.3,
bias_initial_min=-1,
bias_initial_max=1,
bias_min=-15,
bias_max=15,
compatibility_factor_disjoint_genes=1.0,
compatibility_factor_matching_genes=0.5,
probability_mutate_add_connection=0.5,
probability_mutate_add_node=0.2,
compatibility_genome_size_threshold=0)
# Progressbar size
self.progressbar_max = config.population_size
# Create random seed, if none is specified
if seed is None:
seed = np.random.RandomState().randint(2**24)
logger.info("Used Seed: {}".format(seed))
# Create the challenge
self.challenge = BreakoutChallenge()
# Start evaluation
optimizer.evaluate(amount_input_nodes=128,
amount_output_nodes=2,
activation_function=sigmoid_activation,
challenge=self.challenge,
config=config,
seed=seed)
def setUp(self) -> None:
self.config = NeatConfig(population_size=10)
self.generation = gs.create_initial_generation(
3, 2, step_activation, InnovationNumberGeneratorSingleCore(),
SpeciesIDGeneratorSingleCore(), AgentIDGeneratorSingleCore(),
self.config, 1)
for i, agent in zip(range(len(self.generation.agents)),
self.generation.agents):
agent.fitness = i
def test_create_initial_generation(self):
config = NeatConfig(population_size=150,
connection_max_weight=10,
connection_min_weight=-10)
generation1 = gs.create_initial_generation(
10,
3,
step_activation,
InnovationNumberGeneratorSingleCore(),
SpeciesIDGeneratorSingleCore(),
AgentIDGeneratorSingleCore(),
config,
seed=1)
generation2 = gs.create_initial_generation(
10,
3,
step_activation,
InnovationNumberGeneratorSingleCore(),
SpeciesIDGeneratorSingleCore(),
AgentIDGeneratorSingleCore(),
config,
seed=1)
self.assertEqual(0, generation1.number)
self.assertEqual(0, generation2.number)
self.assertEqual(150, len(generation1.agents))
# Compare if generations and generated genomes are the same
for agent1, agent2 in zip(generation1.agents, generation2.agents):
self.assertEqual(agent1.genome.seed, agent2.genome.seed)
for node1, node2, i in zip(agent1.genome.nodes,
agent2.genome.nodes,
range(len(agent1.genome.nodes))):
self.assertEqual(node1.innovation_number,
node2.innovation_number)
self.assertEqual(i, node1.innovation_number)
self.assertEqual(node1.bias, node2.bias)
if node1.node_type != NodeType.INPUT:
self.assertNotAlmostEqual(0,
node1.bias,
delta=0.000000000001)
# Check connection weights
for connection1, connection2, i in zip(
agent1.genome.connections, agent2.genome.connections,
range(len(agent1.genome.connections))):
self.assertEqual(connection1.weight, connection2.weight)
self.assertEqual(connection1.innovation_number,
connection2.innovation_number)
self.assertEqual(i, connection1.innovation_number)
self.assertNotAlmostEqual(0,
connection1.weight,
delta=0.000000000001)
def test_agent(self):
genome = create_genome_structure(5, 2, modified_sigmoid_activation,
NeatConfig(),
InnovationNumberGeneratorSingleCore())
agent = Agent(1, genome)
self.assertEqual(1, agent.id)
self.assertEqual(genome, agent.genome)
self.assertIsNone(agent.neural_network)
self.assertEqual(0, agent.fitness)
self.assertEqual(0, agent.adjusted_fitness)
def setUp(self) -> None:
self.config = NeatConfig(population_size=7)
self.inno_num_generator = InnovationNumberGeneratorSingleCore()
self.species_id_generator = SpeciesIDGeneratorSingleCore()
self.agent_id_generator = AgentIDGeneratorSingleCore()
self.genome1 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome2 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome3 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome4 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome5 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome6 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome7 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.agent1 = Agent(1, self.genome1)
self.agent1.fitness = 1
self.agent2 = Agent(2, self.genome2)
self.agent2.fitness = 2
self.agent3 = Agent(3, self.genome3)
self.agent3.fitness = 3
self.agent4 = Agent(4, self.genome4)
self.agent4.fitness = 4
self.agent5 = Agent(5, self.genome5)
self.agent5.fitness = 5
self.agent6 = Agent(6, self.genome6)
self.agent6.fitness = 6
self.agent7 = Agent(7, self.genome7)
self.agent7.fitness = 7
self.species1 = Species(
self.species_id_generator.get_species_id(), self.agent1.genome,
[self.agent1, self.agent1, self.agent2, self.agent3])
self.species2 = Species(self.species_id_generator.get_species_id(),
self.agent4.genome, [self.agent4, self.agent5])
self.species3 = Species(self.species_id_generator.get_species_id(),
self.agent6.genome, [self.agent6, self.agent7])
def test_set_new_genome_weights(self):
original_genome = Genome(123,
[Node(1, NodeType.INPUT, 1.1, step_activation, 0)],
[Connection(124, 10, 20, 1.2, True),
Connection("124124", 12, 22, 0.8, False)])
config = NeatConfig(connection_initial_min_weight=-10, connection_initial_max_weight=10)
new_genome = set_new_genome_weights(original_genome, np.random.RandomState(2), config=config)
# First 3 random values
# 1. -1.2801019571599248
# 2. -9.481475363442174
# 3. -1.293552147634463
self.assertAlmostEqual(-1.2801019571599248, new_genome.connections[0].weight, delta=0.00000001)
self.assertAlmostEqual(-9.481475363442174, new_genome.connections[1].weight, delta=0.00000001)
def test_create_initial_generation_genome(self):
genome = Genome(20, [
Node("abc", NodeType.INPUT, 0.3, step_activation, 0),
Node("def", NodeType.OUTPUT, 0.4, step_activation, 1)
], [
Connection("x", "abc", "def", 1.0, True),
Connection("y", "def", "abc", -5, True)
])
generation = gs.create_initial_generation_genome(
genome,
InnovationNumberGeneratorSingleCore(),
SpeciesIDGeneratorSingleCore(),
AgentIDGeneratorSingleCore(),
NeatConfig(population_size=3),
seed=1)
self.assertEqual(3, len(generation.agents))
self.assertEqual(0, generation.number)
self.assertEqual(1, len(generation.species_list))
self.assertEqual(3, len(generation.species_list[0].members))
self.assertEqual(1, generation.seed)
# First three random numbers
# 1 - 12710949
# 2 - 4686059
# 3 - 6762380
self.assertEqual(12710949, generation.agents[0].genome.seed)
self.assertEqual(4686059, generation.agents[1].genome.seed)
self.assertEqual(6762380, generation.agents[2].genome.seed)
for node1, node2, node3, i in zip(generation.agents[0].genome.nodes,
generation.agents[1].genome.nodes,
generation.agents[2].genome.nodes,
range(3)):
self.assertEqual(node1.innovation_number, node2.innovation_number)
self.assertEqual(node2.innovation_number, node3.innovation_number)
self.assertEqual(node3.innovation_number, i)
for connection1, connection2, connection3, i in zip(
generation.agents[0].genome.connections,
generation.agents[1].genome.connections,
generation.agents[2].genome.connections, range(3)):
self.assertEqual(connection1.innovation_number,
connection2.innovation_number)
self.assertEqual(connection2.innovation_number,
connection3.innovation_number)
self.assertEqual(connection3.innovation_number, i)
def test_create_genome_structure(self):
config = NeatConfig()
input_nodes = 100
output_nodes = 50
generated_structure = gs.create_genome_structure(
amount_input_nodes=input_nodes,
amount_output_nodes=output_nodes,
activation_function=step_activation,
config=config,
generator=InnovationNumberGeneratorSingleCore())
self.assertEqual(input_nodes + output_nodes,
len(generated_structure.nodes))
self.assertEqual(input_nodes * output_nodes,
len(generated_structure.connections))
for node, i in zip(generated_structure.nodes,
range(input_nodes + output_nodes)):
self.assertEqual(i, node.innovation_number)
self.assertEqual(step_activation, node.activation_function)
self.assertEqual(0, node.bias)
if i < input_nodes:
self.assertEqual(NodeType.INPUT,
node.node_type,
msg="Value i={}".format(i))
self.assertEqual(0, node.x_position)
else:
self.assertEqual(NodeType.OUTPUT,
node.node_type,
msg="Value i={}".format(i))
self.assertEqual(1, node.x_position)
for connection, i in zip(generated_structure.connections,
range(input_nodes * output_nodes)):
self.assertEqual(i, connection.innovation_number)
self.assertEqual(0, connection.weight)
def test_mutate_weights_uniform(self):
config = NeatConfig(probability_weight_mutation=0.6,
probability_random_weight_mutation=0.5,
connection_min_weight=-4,
connection_max_weight=4,
connection_initial_min_weight=-3,
connection_initial_max_weight=3,
weight_mutation_type="uniform",
weight_mutation_uniform_max_change=1)
# Random values for with seed 1
# First connection
# rnd.uniform(0, 1) = 0.417022004702574 -> Mutate yes
# rnd.uniform(0, 1) = 0.7203244934421581 -> Perturb weight
# rnd.uniform(-1, 1) = -0.9997712503653102 -> Value to be subtracted (and clamped)
# Second connection
# rnd.uniform(0, 1) = 0.30233257263183977 -> Mutate yes
# rnd.uniform(0, 1) = 0.14675589081711304 -> Random weight
# rnd.uniform(-3, 3) = -2.445968431387213 -> new random weight
# Third connection
# rnd.uniform(0, 1) = 0.1862602113776709 -> Mutate yes
# rnd.uniform(0, 1) = 0.34556072704304774 -> Random weight
# rnd.uniform(-3, 3) = -0.6193951546159804 -> new random weight
# Fourth connection
# rnd.uniform(0, 1) = 0.538816734003357 -> Mutate yes
# rnd.uniform(0, 1) = 0.4191945144032948 -> Random weight
# rnd.uniform(-3, 3) = 1.1113170023805568 -> new random weight
con1_expected_weight = self.connection1.weight - 0.9997712503653102
new_genome = mutate_weights(self.genome, self.rnd, config)
# Same object
self.assertEqual(self.genome, new_genome)
self.assertAlmostEqual(con1_expected_weight, self.connection1.weight, delta=0.000000000001)
self.assertAlmostEqual(-2.445968431387213, self.connection2.weight, delta=0.000000000001)
self.assertAlmostEqual(-0.6193951546159804, self.connection3.weight, delta=0.000000000001)
self.assertAlmostEqual(1.1113170023805568, self.connection4.weight, delta=0.000000000001)
def test_mutate_weights_normal(self):
config = NeatConfig(probability_weight_mutation=0.6,
probability_random_weight_mutation=0.5,
connection_min_weight=-3,
connection_max_weight=3,
weight_mutation_type="normal",
weight_mutation_normal_sigma=0.5)
# Random values for with seed 1
# First connection
# rnd.uniform(0, 1) = 0.417022004702574 -> Mutate yes
# rnd.uniform(0, 1) = 0.7203244934421581 -> Perturb weight
# rnd.normal(scale=0.5) = -0.26408587613172785 -> Value to be subtracted (and clamped)
# Second connection
# rnd.uniform(0, 1) = 0.39676747423066994 -> Mutate yes
# rnd.uniform(0, 1) = 0.538816734003357 -> Perturb weight
# rnd.normal(scale=0.5) = -0.5364843110780853 -> Value to be subtracted (and clamped)
# Third connection
# rnd.uniform(0, 1) = 0.4191945144032948 -> Mutate yes
# rnd.uniform(0, 1) = 0.6852195003967595 -> Perturb weight
# rnd.normal(scale=0.5) = 0.15951954802854929 -> Value to be added (and clamped)
# Fourth connection
# rnd.uniform(0, 1) = 0.027387593197926163 -> Mutate yes
# rnd.uniform(0, 1) = 0.6704675101784022 -> Perturb weight
# rnd.normal(scale=0.5) = -0.12468518773870504 -> Value to be subtracted (and clamped)
connection1_new_weight = self.connection1.weight - 0.26408587613172785
connection2_new_weight = self.connection2.weight - 0.5364843110780853
connection3_new_weight = self.connection3.weight + 0.15951954802854929
connection4_new_weight = self.connection4.weight - 0.12468518773870504
new_genome = mutate_weights(self.genome, self.rnd, config)
# Same object
self.assertEqual(self.genome, new_genome)
self.assertAlmostEqual(connection1_new_weight, self.connection1.weight, delta=0.000000000001)
self.assertAlmostEqual(connection2_new_weight, self.connection2.weight, delta=0.000000000001)
self.assertAlmostEqual(connection3_new_weight, self.connection3.weight, delta=0.000000000001)
self.assertAlmostEqual(connection4_new_weight, self.connection4.weight, delta=0.000000000001)
def test_mutate_bias_uniform(self):
config = NeatConfig(probability_bias_mutation=0.6,
bias_max=3,
bias_min=-3,
bias_initial_min=-2,
bias_initial_max=2,
probability_random_bias_mutation=0.5,
bias_mutation_uniform_max_change=1,
bias_mutation_type="uniform")
# rnd.uniform(0, 1) = 0.417022004702574 -> Mutate
# rnd.uniform(0, 1) = 0.7203244934421581 -> Perturb
# rnd.uniform(-1, 1) = -0.9997712503653102 -> Substract form bias
# rnd.uniform(0, 1) = 0.30233257263183977 -> Mutate
# rnd.uniform(0, 1) = 0.14675589081711304 -> Random weight
# rnd.uniform(-2, 2) = -1.6306456209248088 -> new random weight
old_output_bias = self.node_output1.bias
new_genome = mutate_bias(self.genome, self.rnd, config)
self.assertAlmostEqual(old_output_bias - 0.9997712503653102, new_genome.nodes[2].bias, delta=0.000000001)
self.assertAlmostEqual(-1.6306456209248088, new_genome.nodes[3].bias, delta=0.0000000001)
def test_mutate_bias_normal(self):
config = NeatConfig(probability_bias_mutation=0.6,
bias_max=3,
bias_min=-3,
bias_initial_min=-2,
bias_initial_max=2,
probability_random_bias_mutation=0.6,
bias_mutation_normal_sigma=1.0,
bias_mutation_type="normal")
# rnd.uniform(0, 1) = 0.417022004702574 -> Mutate
# rnd.uniform(0, 1) = 0.7203244934421581 -> Perturb
# rnd.normal(scale=1) = -0.5281717522634557 -> Substract form bias
# rnd.uniform(0, 1) = 0.39676747423066994 -> Mutate
# rnd.uniform(0, 1) = 0.538816734003357 -> Random weight
# rnd.uniform(-2, 2) = -0.3232219423868208 -> new random weight
old_output_bias = self.node_output1.bias
new_genome = mutate_bias(self.genome, self.rnd, config)
self.assertAlmostEqual(old_output_bias - 0.5281717522634557, new_genome.nodes[2].bias, delta=0.000000001)
self.assertAlmostEqual(-0.3232219423868208, new_genome.nodes[3].bias, delta=0.0000000001)
def test_build_generation_from_genome(self):
initial_genome = Genome(20, [
Node(1, NodeType.INPUT, 0, step_activation, 0),
Node(2, NodeType.OUTPUT, 0.4, step_activation, 1)
], [Connection(1, 1, 2, 1.0, True),
Connection(2, 1, 2, -5, True)])
rnd = np.random.RandomState(1)
generation = gs._build_generation_from_genome(
initial_genome, SpeciesIDGeneratorSingleCore(),
AgentIDGeneratorSingleCore(), 19, rnd,
NeatConfig(population_size=3))
self.assertEqual(3, len(generation.agents))
self.assertEqual(0, generation.number)
self.assertEqual(1, len(generation.species_list))
self.assertEqual(3, len(generation.species_list[0].members))
self.assertEqual(19, generation.seed)
# First three random numbers
# 1 - 12710949
# 2 - 4686059
# 3 - 6762380
self.assertEqual(12710949, generation.agents[0].genome.seed)
self.assertEqual(4686059, generation.agents[1].genome.seed)
self.assertEqual(6762380, generation.agents[2].genome.seed)
# Check if every weight and bias is not equal to 0
for agent, i in zip(generation.agents, range(len(generation.agents))):
genome = agent.genome
self.assertEqual(i, agent.id)
for node in genome.nodes:
# Input nodes can have bias 0
if node.node_type == NodeType.INPUT:
continue
self.assertNotAlmostEqual(0, node.bias, delta=0.00000000001)
for conn in genome.connections:
self.assertNotAlmostEqual(0, conn.weight, delta=0.00000000001)
def evaluate_fix_structure(self,
optimizer: NeatOptimizer,
seed: int = None):
optimizer.register_callback(self)
config = NeatConfig(allow_recurrent_connections=False,
population_size=150,
compatibility_threshold=3,
connection_min_weight=-5,
connection_max_weight=5,
bias_min=-5,
bias_max=5,
probability_mutate_add_node=0,
probability_mutate_add_connection=0)
# Create random seed, if none is specified
if seed is None:
seed = np.random.RandomState().randint(2**24)
logger.info("Used Seed: {}".format(seed))
genome = Genome(0, [
Node(0, NodeType.INPUT, 0, modified_sigmoid_activation, 0),
Node(1, NodeType.INPUT, 0, modified_sigmoid_activation, 0),
Node(2, NodeType.OUTPUT, 0, modified_sigmoid_activation, 1),
Node(3, NodeType.HIDDEN, 0, modified_sigmoid_activation, 0.5),
Node(4, NodeType.HIDDEN, 0, modified_sigmoid_activation, 0.5)
], [
Connection(1, 0, 3, 0.1, True),
Connection(2, 1, 3, 0.1, True),
Connection(3, 0, 4, 0.1, True),
Connection(4, 1, 4, 0.1, True),
Connection(5, 3, 2, 0.1, True),
Connection(6, 4, 2, 0.1, True)
])
optimizer.evaluate_genome_structure(genome_structure=genome,
challenge=ChallengeXOR(),
config=config,
seed=seed)
def test_randomize_weight_bias(self):
genome = Genome(20, [
Node(1, NodeType.INPUT, 0.0, step_activation, 0),
Node(2, NodeType.OUTPUT, 0.0, step_activation, 1)
], [Connection(3, 1, 2, 0, True),
Connection(4, 2, 1, 0, True)])
config = NeatConfig(bias_initial_min=1,
bias_initial_max=2,
connection_initial_min_weight=3,
connection_initial_max_weight=4)
randomized_genome = gs._randomize_weight_bias(genome,
np.random.RandomState(1),
config)
for node in randomized_genome.nodes:
if node.node_type == NodeType.INPUT:
self.assertEqual(0, node.bias)
else:
self.assertTrue(1 <= node.bias <= 2)
for conn in randomized_genome.connections:
self.assertTrue(3 <= conn.weight <= 4)
def setUp(self) -> None:
self.config = NeatConfig(compatibility_factor_matching_genes=1,
compatibility_factor_disjoint_genes=2,
compatibility_threshold=2.5,
compatibility_genome_size_threshold=0,
population_size=8)
self.g1_nodes = [
Node(1, NodeType.INPUT, 0, step_activation, 0),
Node(2, NodeType.INPUT, 0, step_activation, 0),
Node(3, NodeType.OUTPUT, 1.2, step_activation, 1),
Node(4, NodeType.HIDDEN, 1.5, step_activation, 0.5),
Node(6, NodeType.HIDDEN, 0.5, step_activation, 0.5),
Node(7, NodeType.HIDDEN, 0.2, step_activation, 0.25)
]
self.g1_connections = [
Connection(1, 1, 3, 1.2, True),
Connection(2, 2, 3, 0.5, False),
Connection(3, 1, 4, -1.2, True),
Connection(4, 4, 3, 0.2, True),
Connection(5, 2, 6, 2.0, True),
Connection(6, 6, 3, -1.1, False)
]
self.g1 = Genome(1, self.g1_nodes, self.g1_connections)
self.g2_nodes = [
Node(1, NodeType.INPUT, 0, step_activation, 0),
Node(2, NodeType.INPUT, 0, step_activation, 0),
Node(3, NodeType.OUTPUT, 0.2, step_activation, 1),
Node(4, NodeType.HIDDEN, 1.2, step_activation, 0.5),
Node(5, NodeType.HIDDEN, 2.8, step_activation, 0.5)
]
self.g2_connections = [
Connection(1, 1, 3, 0.8, True),
Connection(2, 2, 3, 1.5, True),
Connection(3, 1, 4, 1.2, True),
Connection(4, 4, 3, 3.2, True),
Connection(6, 6, 3, -1.1, False),
Connection(7, 6, 3, -0.1, False),
Connection(8, 1, 4, -1.1, False)
]
self.g2 = Genome(2, self.g2_nodes, self.g2_connections)
self.agent1 = Agent(1, self.g1)
self.agent1.fitness = 1
self.agent2 = Agent(2, self.g2)
self.agent2.fitness = 2
# Add some more agents, and complete species
self.inno_num_generator = InnovationNumberGeneratorSingleCore()
self.species_id_generator = SpeciesIDGeneratorSingleCore()
self.genome3 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome4 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome5 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome6 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome7 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.genome8 = gs.create_genome_structure(2, 1, step_activation,
self.config,
self.inno_num_generator)
self.agent3 = Agent(3, self.genome3)
self.agent3.fitness = 3
self.agent4 = Agent(4, self.genome4)
self.agent4.fitness = 4
self.agent5 = Agent(5, self.genome5)
self.agent5.fitness = 5
self.agent6 = Agent(6, self.genome6)
self.agent6.fitness = 6
self.agent7 = Agent(7, self.genome7)
self.agent7.fitness = 7
self.agent8 = Agent(8, self.genome8)
self.agent8.fitness = 8
self.species1 = Species(self.species_id_generator.get_species_id(),
self.agent1.genome,
[self.agent1, self.agent2, self.agent3],
max_species_fitness=1.5,
generation_max_species_fitness=10)
self.species2 = Species(self.species_id_generator.get_species_id(),
self.agent4.genome,
[self.agent4, self.agent5, self.agent6],
max_species_fitness=7,
generation_max_species_fitness=5)
self.species3 = Species(self.species_id_generator.get_species_id(),
self.agent6.genome, [self.agent7, self.agent8],
max_species_fitness=0,
generation_max_species_fitness=6)
self.generation = Generation(
21,
2,
agents=[
self.agent1, self.agent2, self.agent3, self.agent4,
self.agent5, self.agent6, self.agent7, self.agent8
],
species_list=[self.species1, self.species2, self.species3])
def evaluate(self, optimizer: NeatOptimizer, seed: int = None, **kwargs):
time = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
# Create reporter
self.fitness_reporter = FitnessReporter()
self.species_reporter = SpeciesReporter()
self.time_reporter = TimeReporter()
self.check_point_reporter = CheckPointReporter(
"/tmp/xor/{}/".format(time), "xor_genome_", lambda _: False)
# Register this class as callback and reporter
optimizer.register_callback(self)
optimizer.register_reporters(self.time_reporter, self.fitness_reporter,
self.species_reporter,
self.check_point_reporter)
config = NeatConfig(allow_recurrent_connections=False,
population_size=150,
compatibility_threshold=3,
connection_min_weight=-15,
connection_max_weight=15,
bias_min=-15,
bias_max=15,
compatibility_factor_disjoint_genes=1.0,
compatibility_factor_matching_genes=0.5,
probability_mutate_add_connection=0.5,
probability_mutate_add_node=0.2)
# config = NeatConfig(allow_recurrent_connections=False,
# population_size=150,
# compatibility_threshold=3,
# connection_min_weight=-15,
# connection_max_weight=15,
# bias_min=-15,
# bias_max=15,
# compatibility_factor_disjoint_genes=1.0,
# compatibility_factor_matching_genes=0.4,
# probability_mutate_add_connection=0.05,
# probability_mutate_add_node=0.03)
# Specify the config
# config = NeatConfig(allow_recurrent_connections=False,
# population_size=400,
# compatibility_threshold=3,
# weight_mutation_type="normal",
# weight_mutation_normal_sigma=1.3,
# connection_initial_min_weight=-5,
# connection_initial_max_weight=5,
# connection_min_weight=-5,
# connection_max_weight=5,
# bias_mutation_type="normal",
# bias_mutation_normal_sigma=1.3,
# bias_initial_min=-1,
# bias_initial_max=1,
# bias_min=-5,
# bias_max=5,
# compatibility_factor_disjoint_genes=1.0,
# compatibility_factor_matching_genes=0.4,
# probability_mutate_add_connection=0.05,
# probability_mutate_add_node=0.03,
# compatibility_genome_size_threshold=10)
# Create random seed, if none is specified
if seed is None:
seed = np.random.RandomState().randint(2**24)
logger.info("Used Seed: {}".format(seed))
optimizer.evaluate(amount_input_nodes=2,
amount_output_nodes=1,
activation_function=tanh_activation,
challenge=ChallengeXOR(),
config=config,
seed=seed)
def setUp(self) -> None:
self.config = NeatConfig(population_size=10)