parser = argparse.ArgumentParser(description='A3C')
parser.add_argument('--env-name',
default='SpaceInvaders-v0',
help='environment to train on (default: SpaceInvaders-v0)')
parser.add_argument('--logs-path',
default="./log.txt",
help='path to logs (default: `./log.txt`)')
parser.add_argument(
'--pretrained-weights',
default=None,
help='path to pretrained weights (default: None – evaluate random model)')
if __name__ == '__main__':
cmd_args = parser.parse_args()
logging.basicConfig(filename=cmd_args.logs_path, level=logging.INFO)
env = make_atari(cmd_args.env_name, clip=False)
model = ActorCritic(env.observation_space.shape[0], env.action_space.n)
if cmd_args.pretrained_weights is not None:
model.load_weights(cmd_args.pretrained_weights)
else:
print(
"You have not specified path to model weigths, random plays will be performed"
)
model.eval()
results = record_video(model, env)
log_message = "evaluated on pretrained weights: {}, results: {}".format(
cmd_args.pretrained_weights, results)
print(log_message)
logging.info(log_message)
default=None,
help=
'path to pretrained weights and optimizer params (default: if None – train from scratch)'
)
if __name__ == '__main__':
cmd_args = parser.parse_args()
config = Config.fromYamlFile('config.yaml')
args = config.train
args.__dict__.update(vars(cmd_args))
env = make_atari(args.env_name)
shared_model = ActorCritic(env.observation_space.shape, env.action_space.n)
if args.pretrained_weights is not None:
shared_model.load_weights(args.pretrained_weights)
shared_model.share_memory()
optimizer = SharedAdam(shared_model.parameters(), lr=args.learning_rate)
if args.pretrained_weights is not None:
optimizer.load_params(
args.pretrained_weights.replace('weights/', 'optimizer_params/'))
optimizer.share_memory()
processes = []
lock = mp.Lock()
total_steps = Value('i', 0)
p = mp.Process(target=test_worker,
args=(args, shared_model, total_steps, optimizer))