def codeepneat_xor_example(_):
"""
This Example evolves a CoDeepNEAT population on the XOR problem for 100 generations, using fixed speciation for
the modules and blueprints. Subsequently the best genome is trained for a final 100 epochs and its genotype and
Tensorflow model are backed up.
"""
# Set standard configuration specific to TFNE but not the neuroevolution process
logging_level = logging.INFO
config_file_path = './codeepneat_xor_fixed_example_config.cfg'
backup_dir_path = './tfne_state_backups/'
max_generations = 100
max_fitness = None
# Read in optionally supplied flags, changing the just set standard configuration
if flags.FLAGS.logging_level is not None:
logging_level = flags.FLAGS.logging_level
if flags.FLAGS.config_file is not None:
config_file_path = flags.FLAGS.config_file
if flags.FLAGS.backup_dir is not None:
backup_dir_path = flags.FLAGS.backup_dir
if flags.FLAGS.max_generations is not None:
max_generations = flags.FLAGS.max_generations
if flags.FLAGS.max_fitness is not None:
max_fitness = flags.FLAGS.max_fitness
# Set logging, parse config
logging.set_verbosity(logging_level)
config = tfne.parse_configuration(config_file_path)
# Initialize the environment and the specific NE algorithm
environment = tfne.environments.XOREnvironment(weight_training=True,
config=config,
verbosity=logging_level)
ne_algorithm = tfne.algorithms.CoDeepNEAT(config)
# Initialize evolution engine and supply config as well as initialized NE algorithm and evaluation environment.
engine = tfne.EvolutionEngine(ne_algorithm=ne_algorithm,
environment=environment,
backup_dir_path=backup_dir_path,
max_generations=max_generations,
max_fitness=max_fitness)
# Start training process, returning the best genome when training ends
best_genome = engine.train()
print("Best genome returned by evolution:\n")
print(best_genome)
# Increase epoch count in environment for a final training of the best genome. Train the genome and then replay it.
print("Training best genome for 100 epochs...\n")
environment.epochs = 100
environment.eval_genome_fitness(best_genome)
environment.replay_genome(best_genome)
# Serialize and save genotype and Tensorflow model to demonstrate serialization
best_genome.save_genotype(save_dir_path='./best_genome_genotype/')
best_genome.save_model(file_path='./best_genome_model/')
def codeepneat_xor_example(_):
""""""
# Set standard configuration specific to TFNE but not the neuroevolution process
logging_level = logging.INFO
config_file_path = './codeepneat_xor_example_config.cfg'
backup_dir_path = './population_backups/'
num_cpus = None
num_gpus = None
max_generations = 100
max_fitness = None
# Read in optionally supplied flags, changing the just set standard configuration
if flags.FLAGS.logging_level is not None:
logging_level = flags.FLAGS.logging_level
if flags.FLAGS.config_file is not None:
config_file_path = flags.FLAGS.config_file
if flags.FLAGS.backup_dir is not None:
backup_dir_path = flags.FLAGS.backup_dir
if flags.FLAGS.num_cpus is not None:
num_cpus = flags.FLAGS.num_cpus
if flags.FLAGS.num_gpus is not None:
num_gpus = flags.FLAGS.num_gpus
if flags.FLAGS.max_generations is not None:
max_generations = flags.FLAGS.max_generations
if flags.FLAGS.max_fitness is not None:
max_fitness = flags.FLAGS.max_fitness
# Set logging, parse config
logging.set_verbosity(logging_level)
config = tfne.parse_configuration(config_file_path)
# Set (not initialize) the environment and initialize the specific NE algorithm
environment = tfne.environments.XOREnvironment
ne_algorithm = tfne.CoDeepNEAT(config, environment)
# Initialize evolution engine and supply config as well as initialized NE elements
engine = tfne.EvolutionEngine(ne_algorithm=ne_algorithm,
backup_dir_path=backup_dir_path,
num_cpus=num_cpus,
num_gpus=num_gpus,
max_generations=max_generations,
max_fitness=max_fitness)
# Start training process, returning the best genome when training ends
best_genome = engine.train()
# Show string representation of best genome, visualize it and then save it
print("Best Genome returned by evolution:\n")
print(best_genome)
best_genome.save_genotype(save_dir_path='./')
best_genome.save_model(save_dir_path='./')
def test_codeepneat_5():
# Create test config
config = tfne.parse_configuration(
os.path.dirname(__file__) + '/test_codeepneat_4_config.cfg')
environment = tfne.environments.CIFAR10Environment(weight_training=True,
config=config)
ne_algorithm = tfne.algorithms.CoDeepNEAT(config)
# Start test
engine = tfne.EvolutionEngine(ne_algorithm=ne_algorithm,
environment=environment,
backup_dir_path=tempfile.gettempdir(),
max_generations=2,
max_fitness=None)
engine.train()
# Sanity check state of the algorithm
sanity_check_algorithm_state(ne_algorithm)