help="Almacena videos y estados de la performance del agente",
action="store_true",
default=False)
args.add_argument(
"--output-dir",
help=
"Directorio para almacenar los outputs. Por defecto = ./trained_models/results",
default="./trained_models/results")
args = args.parse_args()
manager = ParamsManager(args.params_file)
summary_filename_prefix = manager.get_agent_params()['summary_filename_prefix']
summary_filename = summary_filename_prefix + args.env + datetime.now(
).strftime("%y-%m-%d-%H-%M")
writer = SummaryWriter(summary_filename)
manager.export_agent_params(summary_filename + "/" + "agent_params.json")
manager.export_environment_params(summary_filename + "/" +
"environment_params.json")
global_step_num = 0
use_cuda = manager.get_agent_params()['use_cuda']
device = torch.device(
"cuda:" +
str(args.gpu_id) if torch.cuda.is_available() and use_cuda else "cpu")
seed = manager.get_agent_params()['seed']
torch.manual_seed(seed)
np.random.seed(seed)
if torch.cuda.is_available() and use_cuda:
torch.cuda.manual_seed_all(seed)
class DeepQLearner(object):
type=int,
default=0,
metavar="GPU_ID")
args = parser.parse_args()
global_step_num = 0
params_manager = ParamsManager(args.params_file)
agent_algo = "A2C"
summary_file_path_prefix = params_manager.get_agent_params(
)['summary_file_path_prefix']
summary_file_path = summary_file_path_prefix + agent_algo + '_' + args.env + "_" + datetime.now(
).strftime("%y-%m-%d-%H-%M")
writer = SummaryWriter(summary_file_path)
# Export the parameters as json files to the log directory to keep track of the parameters used in each experiment
params_manager.export_env_params(summary_file_path + "/" + "env_params.json")
params_manager.export_agent_params(summary_file_path + "/" +
"agent_params.json")
use_cuda = params_manager.get_agent_params()['use_cuda']
# Introduced in PyTorch 0.4
device = torch.device(
"cuda:" +
str(args.gpu_id) if torch.cuda.is_available() and use_cuda else "cpu")
seed = params_manager.get_agent_params()[
'seed'] # With the intent to make the results reproducible
torch.manual_seed(seed)
np.random.seed(seed)
if torch.cuda.is_available() and use_cuda:
torch.cuda.manual_seed_all(seed)
Transition = namedtuple("Transition", ["s", "value_s", "a", "log_prob_a"])
# Parameter Manager
params_manager = ParamsManager(args.params_file)
seed = params_manager.get_agent_params()['seed']
summary_file_path_prefix = params_manager.get_agent_params(
)['summary_file_path_prefix']
summary_file_path = summary_file_path_prefix + args.env + \
'_' + datetime.now().strftime('%y-%m-%d-%H-%M')
if not exists(summary_file_path):
makedirs(summary_file_path)
writer = SummaryWriter(summary_file_path)
params_manager.export_env_params(join(summary_file_path, 'env_params.json'))
params_manager.export_agent_params(join(summary_file_path,
'agent_params.json'))
global_step_num = 0
# GPU Setting
use_cuda = params_manager.get_agent_params()['use_cuda']
device = torch.device(
'cuda:' +
str(args.gpu_id) if torch.cuda.is_available() and use_cuda else 'cpu')
torch.manual_seed(seed)
np.random.seed(seed)
if torch.cuda.is_available() and use_cuda:
torch.cuda.manual_seed_all(seed)
if __name__ == "__main__":
env_conf = params_manager.get_env_params()