parser.add_argument('--net_size', type=int, default=None)
parser.add_argument('--exp_name', type=str, default=None)
# parser.add_argument('--exp_name', type=str, default=timestamp())
# parser.add_argument('--mode', type=str, default='local')
# parser.add_argument('--log_dir', type=str, default=None)
args = parser.parse_args()
return args
if __name__ == "__main__":
args = parse_args()
if args.env is None:
variant = ENV_PARAMS[DEFAULT_ENV]
else:
variant = ENV_PARAMS[args.env]
# Net size
if args.net_size is not None:
variant['net_size'] = args.net_size
# Experiment name
if args.exp_name is None:
exp_name = variant['env_name']
else:
exp_name = args.exp_name
setup_logger(exp_name, variant=variant)
experiment(variant)
expt_variant['env_params']['subtask'] = args.subtask
expt_variant['log_dir'] = args.log_dir
expt_variant['load_dir'] = args.load_dir
# Net size
if args.net_size is not None:
expt_variant['net_size'] = args.net_size
expt_variant['gpu'] = args.gpu
expt_variant['seed'] = args.seed
expt_variant['render_q'] = args.render_q
# Algo params
expt_variant['algo_params']['render'] = args.render
log_dir = setup_logger(expt_name,
variant=expt_variant,
snapshot_mode=args.snap_mode,
snapshot_gap=args.snap_gap,
log_dir=args.log_dir)
dir_filename = os.path.realpath(__file__)
filename = os.path.split(dir_filename)[1]
copyfile(dir_filename, os.path.join(log_dir, filename))
algo = experiment(expt_variant)
# input('Press a key to close the script...')
)
algorithm = SoftActorCritic(env=env,
policy=policy,
qf=qf,
vf=vf,
**variant['algo_params'])
if ptu.gpu_enabled():
algorithm.cuda()
algorithm.train()
if __name__ == "__main__":
# noinspection PyTypeChecker
variant = dict(
algo_params=dict(
num_epochs=1000,
num_steps_per_epoch=1000,
num_steps_per_eval=1000,
batch_size=128,
max_path_length=999,
discount=0.99,
soft_target_tau=0.001,
policy_lr=3E-4,
qf_lr=3E-4,
vf_lr=3E-4,
),
net_size=300,
)
setup_logger('name-of-experiment', variant=variant)
experiment(variant)
