def main():
env = gym.make('SlimeVolley-v0')
env.seed(123)
POPULATION_SIZE = 20
MAX_GENERATION = 100
MUTATION_RATE = 0.4
CROSSOVER_RATE = 0.7
INPUT_SIZE = 12
OUTPUT_SIZE = 3
assert POPULATION_SIZE % 2 == 0
p = Population(DeepMLPTorchIndividual(INPUT_SIZE, 0,
OUTPUT_SIZE), POPULATION_SIZE,
MAX_GENERATION, MUTATION_RATE, CROSSOVER_RATE, 0.0)
p.set_population([
DeepMLPTorchIndividual(INPUT_SIZE, 0, OUTPUT_SIZE)
for _ in range(POPULATION_SIZE)
])
p.run(
env,
generation,
verbose=True,
log=True,
output_folder=f'model-layers={INPUT_SIZE}-{HIDDEN_SIZE}-{OUTPUT_SIZE}',
save_as_pytorch=False)
if child1.fitness + child2.fitness > parent1.fitness + parent2.fitness:
new_population[i] = child1
new_population[i + 1] = child2
else:
new_population[i] = parent1
new_population[i + 1] = parent2
if __name__ == '__main__':
env = gym.make('CarRacing-v0')
POPULATION_SIZE = 100
MAX_GENERATION = 2
MUTATION_RATE = 0.1
CROSSOVER_RATE = 0.8
p = Population(ConvNetTorchIndividal(None, None, None), POPULATION_SIZE, MAX_GENERATION, MUTATION_RATE, CROSSOVER_RATE, 0)
p.run(env, generation, verbose=False, output_folder='')
env.close()
'''
초기화
for in max_generation (총 세대수){
for old_population : 적합성 계산 ( 100 episode * population_size )
run_generation(env, old, new, mutation, crossover){
for old population / 2
parent1,parent2 = 적합성 기준 2개 select
child1,child2 = (parent1,parent2,crossover){
# 4.
if np.random.rand() < p_inversion:
child1.weights_biases = inversion(child1.weights_biases)
child2.weights_biases = inversion(child2.weights_biases)
new_population.append(child1)
new_population.append(child2)
if __name__ == '__main__':
env = gym.make('BipedalWalker-v2')
env.seed(123)
POPULATION_SIZE = 10
MAX_GENERATION = 20
MUTATION_RATE = 0.01
CROSSOVER_RATE = 0.8
INVERSION_RATE = 0.02
INPUT_SIZE = 10
HIDDEN_SIZE = 16
OUTPUT_SIZE = 4
p = Population(MLPTorchIndividual(INPUT_SIZE, HIDDEN_SIZE, OUTPUT_SIZE),
POPULATION_SIZE, MAX_GENERATION, MUTATION_RATE,
CROSSOVER_RATE, INVERSION_RATE)
p.run(env, generation_new, verbose=True, log=True, output_folder='')
env.close()
if __name__ == '__main__':
env = gym.make('BipedalWalker-v2')
env.seed(123)
POPULATION_SIZE = 10
MAX_GENERATION = 10
MUTATION_RATE = 0.01
CROSSOVER_RATE = 0.8
INPUT_SIZE = 24
OUTPUT_SIZE = 4
assert POPULATION_SIZE % 2 == 0
p = Population(DeepBipedalWalkerIndividual(INPUT_SIZE, 0,
OUTPUT_SIZE), POPULATION_SIZE,
MAX_GENERATION, MUTATION_RATE, CROSSOVER_RATE, 0.0)
p.set_population([
DeepBipedalWalkerIndividual(INPUT_SIZE, 0, OUTPUT_SIZE)
for _ in range(POPULATION_SIZE)
])
p.run(
env,
generation,
verbose=True,
log=True,
output_folder=f'model-layers={INPUT_SIZE}-{HIDDEN_SIZE}-{OUTPUT_SIZE}',
save_as_pytorch=False)
env.close()
new_population[i + 1] = child2
else:
new_population[i] = parent1
new_population[i + 1] = parent2
if __name__ == '__main__':
env = gym.make('CartPole-v1')
env.seed(123)
POPULATION_SIZE = 100
MAX_GENERATION = 20
MUTATION_RATE = 0.4
CROSSOVER_RATE = 0.9
INPUT_SIZE = 4
HIDDEN_SIZE = 2
OUTPUT_SIZE = 1
p = Population(MLPIndividual(INPUT_SIZE, HIDDEN_SIZE,
OUTPUT_SIZE), POPULATION_SIZE, MAX_GENERATION,
MUTATION_RATE, CROSSOVER_RATE, 0)
p.run(env,
generation,
verbose=True,
output_folder='../../models/cartpole',
log=True)
env.close()