def evolve(pop: Population, pop_name: str):
"""Evolve with the set constraints in mind."""
# Create the next generation from the current generation
pop.population = pop.reproduction.reproduce(
config=pop.config,
species=pop.species,
generation=pop.generation,
logger=pop.log,
)
# Constraint each of the population's new genomes to the given topology
for g in pop.population.values():
enforce_topology(pop_name, genome=g)
# Check for complete extinction
if not pop.species.species:
pop.reporters.complete_extinction(logger=pop.log)
# If requested by the user, create a completely new population, otherwise raise an exception
pop.population = pop.reproduction.create_new(
config=pop.config, num_genomes=pop.config.population.pop_size)
# Divide the new population into species
pop.species.speciate(config=pop.config,
population=pop.population,
generation=pop.generation,
logger=pop.log)
# Add to each of the species its elites
pop.update_species_fitness_hist()
# Increment generation count
pop.generation += 1
def main(topology_id: int,
iterations: int,
eval_interval: int,
experiment_id: int,
hops: float = 0.1,
weight_range: float = 1,
mutate_combine: bool = False,
mutate_bias: bool = True,
mutate_reset: bool = True,
mutate_update: bool = True,
mutate_candidate: bool = True,
init_population_size: int = 1000,
unused_cpu: int = 2):
"""
Run the fifth experiment.
:param topology_id: Chosen topology to investigate
:param iterations: Number of training iterations performed
:param eval_interval: After how much training iterations evaluation is performed
:param experiment_id: ID of the experiment used to train and evaluate the population on
:param hops: Hops between variable-configurations
:param weight_range: Range of deviation for one's weights
:param mutate_combine: Combine the best results of each mutation type
:param mutate_bias: Mutate the GRU's bias values
:param mutate_reset: Mutate the reset-gate's weights
:param mutate_update: Mutate the update-gate's weights
:param mutate_candidate: Mutate the candidate-state's weights
:param init_population_size: Initial size of the randomized population
:param unused_cpu: Number of CPU-cores not used
"""
# Get the population
name = f"experiment{experiment_id}_topology{topology_id}_hops{hops}_range{weight_range}"
pop = Population(
name=name,
folder_name='experiment5',
use_backup=False,
)
# Define the games specific to the experiment
_, game_ids_eval = get_game_ids(experiment_id=experiment_id)
# Set the genomes if population is new
if pop.generation == 0:
# Get the requested topology-type
if topology_id == 1:
topology = get_topology1
elif topology_id == 2:
topology = get_topology2
else:
raise Exception(f"Topology {topology_id} not supported.")
# Initialize the population with a randomized population
pop.population = dict()
for gid in range(init_population_size):
new_genome = topology(pop.config, random_init=True)
new_genome.key = gid
pop.population[gid] = new_genome
# Perform an initial training
train_population(pop=pop, games=game_ids_eval, unused_cpu=unused_cpu)
pop.generation = 0 # Don't count initial training as a generation
# Get the fittest genome
best = None
for g in pop.population.values():
if best is None or g.fitness > best.fitness: best = g
pop.best_genome = best
visualize_best_genome(pop=pop)
# Test the initial population
test_population(pop=pop, games=game_ids_eval)
# Set the most fit genome as the starting-point
set_population(pop=pop,
hops=hops,
weight_range=weight_range,
mutate_bias=mutate_bias,
mutate_reset=mutate_reset,
mutate_update=mutate_update,
mutate_candidate=mutate_candidate)
# Evaluate the population
for i in range(iterations):
train_population(pop=pop, games=game_ids_eval, unused_cpu=unused_cpu)
evaluate_fitness(pop=pop,
hops=hops,
weight_range=weight_range,
mutate_combine=mutate_combine,
mutate_bias=mutate_bias,
mutate_reset=mutate_reset,
mutate_update=mutate_update,
mutate_candidate=mutate_candidate)
visualize_best_genome(pop=pop)
if pop.generation % eval_interval == 0:
test_population(pop=pop, games=game_ids_eval)
set_population(pop=pop,
hops=hops,
weight_range=weight_range,
mutate_bias=mutate_bias,
mutate_reset=mutate_reset,
mutate_update=mutate_update,
mutate_candidate=mutate_candidate)
def set_population(
pop: Population,
hops: float,
weight_range: float,
mutate_bias: bool = False,
mutate_reset: bool = False,
mutate_update: bool = False,
mutate_candidate: bool = False,
):
"""Set the given population as mutations of the given genome."""
assert weight_range > hops
r = int(weight_range / hops)
# Re-initialize the weight around the provided genome
pop.population = dict()
genome_key = 0
# Create genome-mutations
if mutate_bias:
for i in range(3):
for a in range(-r, r + 1):
new_genome = copy.deepcopy(pop.best_genome)
new_genome.nodes[2].bias_h[
i] = new_genome.nodes[2].bias_h[i] + a * hops
new_genome.key = genome_key
pop.population[genome_key] = new_genome
genome_key += 1
if mutate_reset:
for a in range(-r, r + 1):
for b in range(-r, r + 1):
new_genome = copy.deepcopy(pop.best_genome)
new_genome.nodes[2].weight_xh_full[
0, 0] = new_genome.nodes[2].weight_xh_full[0, 0] + a * hops
new_genome.nodes[2].weight_hh[
0, 0] = new_genome.nodes[2].weight_hh[0, 0] + b * hops
new_genome.key = genome_key
pop.population[genome_key] = new_genome
genome_key += 1
if mutate_update:
for a in range(-r, r + 1):
for b in range(-r, r + 1):
new_genome = copy.deepcopy(pop.best_genome)
new_genome.nodes[2].weight_xh_full[
1, 0] = new_genome.nodes[2].weight_xh_full[1, 0] + a * hops
new_genome.nodes[2].weight_hh[
1, 0] = new_genome.nodes[2].weight_hh[1, 0] + b * hops
new_genome.key = genome_key
pop.population[genome_key] = new_genome
genome_key += 1
if mutate_candidate:
for a in range(-r, r + 1):
for b in range(-r, r + 1):
new_genome = copy.deepcopy(pop.best_genome)
new_genome.nodes[2].weight_xh_full[
2, 0] = new_genome.nodes[2].weight_xh_full[2, 0] + a * hops
new_genome.nodes[2].weight_hh[
2, 0] = new_genome.nodes[2].weight_hh[2, 0] + b * hops
new_genome.key = genome_key
pop.population[genome_key] = new_genome
genome_key += 1
# Save the updated population
pop.save()
