parser.add_argument('--output', help='output directory for submission', default='output_dir')
parser.add_argument('--episode', help='number of episode to eval', default=100, type=int)
args = parser.parse_args()
ENV_NAME = args.env
logger.info("Environment Name: {}".format(ENV_NAME))
NUM_ACTIONS = get_player().action_space.n
logger.info("Number of actions: {}".format(NUM_ACTIONS))
if args.gpu:
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(PredictConfig(
model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['policy']))
if args.task == 'play':
play_n_episodes(get_player(train=False), pred,
args.episode, render=True)
elif args.task == 'eval':
eval_model_multithread(pred, args.episode, get_player)
elif args.task == 'dump_video':
play_n_episodes(
get_player(train=False, dumpdir=args.output),
pred, args.episode)
else:
train()
NUM_ACTIONS = get_player().action_space.n
logger.info("Number of actions: {}".format(NUM_ACTIONS))
if args.gpu:
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(
PredictConfig(model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['policy']))
if args.task == 'play':
play_n_episodes(get_player(train=False),
pred,
args.episode,
render=True)
elif args.task == 'eval':
eval_model_multithread(pred, args.episode, get_player)
elif args.task == 'gen_submit':
play_n_episodes(get_player(train=False, dumpdir=args.output), pred,
args.episode)
# gym.upload(args.output, api_key='xxx')
else:
dirname = os.path.join('train_log', 'train-atari-{}'.format(ENV_NAME))
logger.set_logger_dir(dirname)
config = get_config()
if args.load:
config.session_init = get_model_loader(args.load)
trainer = SimpleTrainer(
ENV_NAME = args.env
USE_GYM = not ENV_NAME.endswith('.bin')
# set num_actions
num_actions = get_player().action_space.n
logger.info("ENV: {}, Num Actions: {}".format(args.env, num_actions))
state_shape = IMAGE_SIZE + (3, ) if USE_GYM else IMAGE_SIZE
model = Model(state_shape, FRAME_HISTORY, args.algo, num_actions)
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(
PredictConfig(model=model,
session_init=SmartInit(args.load),
input_names=['state'],
output_names=['Qvalue']))
if args.task == 'play':
play_n_episodes(get_player(viz=0.01), pred, 100, render=True)
elif args.task == 'eval':
eval_model_multithread(pred, args.num_eval, get_player)
else:
logger.set_logger_dir(
os.path.join(
'train_log',
'DQN-{}'.format(os.path.basename(args.env).split('.')[0])))
config = get_config(model)
config.session_init = SmartInit(args.load)
launch_train_with_config(config, SimpleTrainer())
session_init=get_model_loader(args.load),
input_names=state_names,
output_names=qvalue_names,
)
)
# demo pretrained model one episode at a time
if args.task == "play":
play_n_episodes(
get_player(
files_list=args.files,
viz=0,
saveGif=args.saveGif,
saveVideo=args.saveVideo,
task="play",
agents=args.agents,
fiducials=args.fiducials,
infDir=args.inferDir,
),
pred,
num_files,
agents=args.agents,
fidname=[fidNumToName[i] for i in args.fiducials],
infDir=args.inferDir,
)
# run episodes in parallel and evaluate pretrained model
elif args.task == "eval":
play_n_episodes(
get_player(
files_list=args.files,
viz=0,
saveGif=args.saveGif,
if args.gpu:
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
if args.task != 'train':
assert args.load is not None
cfg = PredictConfig(model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['policy'])
if args.task == 'play':
play_model(cfg, get_player(viz=0.01))
elif args.task == 'eval':
eval_model_multithread(cfg, args.episode, get_player)
elif args.task == 'gen_submit':
play_n_episodes(get_player(train=False, dumpdir=args.output),
OfflinePredictor(cfg), args.episode)
# gym.upload(output, api_key='xxx')
else:
dirname = os.path.join('train_log', 'train-atari-{}'.format(ENV_NAME))
logger.set_logger_dir(dirname)
nr_gpu = get_nr_gpu()
trainer = QueueInputTrainer
if nr_gpu > 0:
if nr_gpu > 1:
predict_tower = list(range(nr_gpu))[-nr_gpu // 2:]
else:
predict_tower = [0]
PREDICTOR_THREAD = len(predict_tower) * PREDICTOR_THREAD_PER_GPU
train_tower = list(range(nr_gpu))[:-nr_gpu // 2] or [0]
logger.info(
NUM_ACTIONS = get_player().action_space.n
logger.info("Environment: {}, number of actions: {}".format(
ENV_NAME, NUM_ACTIONS))
if args.gpu:
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(
PredictConfig(model=Model(),
session_init=SmartInit(args.load),
input_names=['state'],
output_names=['policy']))
if args.task == 'play':
filename = "/mnt/research/judy/reward_shaping/expert_data/batch_{}.npz".format(
args.save_id)
play_n_episodes(get_player(train=False),
pred,
args.episode,
render=args.render,
save=args.save,
filename=filename)
elif args.task == 'eval':
eval_model_multithread(pred, args.episode, get_player)
elif args.task == 'dump_video':
play_n_episodes(get_player(train=False, dumpdir=args.output), pred,
args.episode)
else:
train()
if args.gpu:
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
ENV_NAME = args.env
# USE_GYM = not ENV_NAME.endswith('.bin')
IMAGE_CHANNEL = 2 if USE_GYM else 2
METHOD = args.algo
# set num_actions
NUM_ACTIONS = get_player().action_space.n
logger.info("ENV: {}, Num Actions: {}".format(ENV_NAME, NUM_ACTIONS))
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(
PredictConfig(model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['Qvalue']))
if args.task == 'play':
play_n_episodes(get_player(viz=0.01), pred, 100)
elif args.task == 'eval':
eval_model_multithread(pred, EVAL_EPISODE, get_player)
else:
logger.set_logger_dir(
os.path.join(
'train_log',
'DQN-{}'.format(os.path.basename(ENV_NAME).split('.')[0])))
config = get_config()
if args.load:
config.session_init = get_model_loader(args.load)
launch_train_with_config(config, SimpleTrainer())
agents=args.agents,
reward_strategy=args.reward_strategy)
NUM_ACTIONS = init_player.action_space.n
num_files = init_player.files.num_files
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(
PredictConfig(model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['Qvalue']))
# demo pretrained model one episode at a time
if args.task == 'play' or args.task == 'eval':
play_n_episodes(
get_player(files_list=args.files,
viz=0.01,
saveGif=args.saveGif,
saveVideo=args.saveVideo,
task=args.task,
agents=args.agents,
reward_strategy=args.reward_strategy), pred,
num_files)
else: # train model
logger_dir = os.path.join(args.logDir, args.name)
logger.set_logger_dir(logger_dir)
config = get_config(args.files)
if args.load: # resume training from a saved checkpoint
config.session_init = get_model_loader(args.load)
launch_train_with_config(config, SimpleTrainer())
choices=['DQN', 'Double', 'Dueling'], default='Double')
args = parser.parse_args()
if args.gpu:
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
ROM_FILE = args.rom
METHOD = args.algo
# set num_actions
NUM_ACTIONS = AtariPlayer(ROM_FILE).action_space.n
logger.info("ROM: {}, Num Actions: {}".format(ROM_FILE, NUM_ACTIONS))
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(PredictConfig(
model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['Qvalue']))
if args.task == 'play':
play_n_episodes(get_player(viz=0.01), pred, 100)
elif args.task == 'eval':
eval_model_multithread(pred, EVAL_EPISODE, get_player)
else:
logger.set_logger_dir(
os.path.join('train_log', 'DQN-{}'.format(
os.path.basename(ROM_FILE).split('.')[0])))
config = get_config()
if args.load:
config.session_init = get_model_loader(args.load)
launch_train_with_config(config, SimpleTrainer())
NUM_ACTIONS = init_player.action_space.n
num_files = init_player.files.num_files
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(
PredictConfig(model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['Qvalue']))
# demo pretrained model one episode at a time
if args.task == 'play':
play_n_episodes(
get_player(files_list=args.files,
viz=0.01,
saveGif=args.saveGif,
saveVideo=args.saveVideo,
task='play'), pred, num_files)
# run episodes in parallel and evaluate pretrained model
elif args.task == 'eval':
play_n_episodes(
get_player(files_list=args.files,
viz=0.01,
saveGif=args.saveGif,
saveVideo=args.saveVideo,
task='eval'), pred, num_files)
else: # train model
logger_dir = os.path.join(args.logDir, args.name)
logger.set_logger_dir(logger_dir)
config = get_config(args.files)
pred = OfflinePredictor(
PredictConfig(
model=Model(agents=args.agents),
session_init=get_model_loader(args.load),
input_names=state_names,
output_names=qvalue_names,
))
# demo pretrained model one episode at a time
if args.task == "play":
play_n_episodes(
get_player(
files_list=args.files,
viz=0.01,
saveGif=args.saveGif,
saveVideo=args.saveVideo,
task="play",
agents=args.agents,
),
pred,
num_files,
agents=args.agents,
)
# run episodes in parallel and evaluate pretrained model
elif args.task == "eval":
play_n_episodes(
get_player(
files_list=args.files,
viz=0,
saveGif=args.saveGif,
saveVideo=args.saveVideo,
task="eval",
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
if args.task != 'train':
assert args.load is not None
cfg = PredictConfig(
model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['policy'])
if args.task == 'play':
play_model(cfg, get_player(connection=(args.ip, args.port), viz=0.01))
# pl = get_player(connection=(args.ip, args.port))
# pl.action(0)
# print(pl.current_state())
elif args.task == 'eval':
eval_model_multithread(cfg, args.episode, get_player)
elif args.task == 'gen_submit':
play_n_episodes(
get_player(connection=(args.ip, args.port), train=False, dumpdir=args.output),
OfflinePredictor(cfg), args.episode)
# gym.upload(output, api_key='xxx')
else:
dirname = os.path.join('train_log', 'train-unity3d-{}'.format(ENV_NAME))
logger.set_logger_dir(dirname)
config = get_config()
if args.load:
config.session_init = get_model_loader(args.load)
trainer = QueueInputTrainer if config.nr_tower == 1 else AsyncMultiGPUTrainer
trainer(config).train()
# IMAGE_CHANNEL = 3 if USE_GYM else 1
# IMAGE_CHANNEL = 5
METHOD = args.algo
# set num_actions
# NUM_ACTIONS = get_player().action_space.n
raw_list, lbl_list = init_data()
NUM_ACTIONS = np.prod(get_player().agent_out_shape)
logger.info("ENV: {}, Num Actions: {}".format(ENV_NAME, NUM_ACTIONS))
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(
PredictConfig(model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['Qvalue']))
if args.task == 'play':
# play_n_episodes(get_player(viz=0.01), pred, 100)
play_n_episodes(get_player(), pred, 100)
elif args.task == 'eval':
eval_model_multithread(pred, EVAL_EPISODE, get_player)
else:
logger.set_logger_dir(
os.path.join(
'train_log', METHOD +
'-DQN-{}'.format(os.path.basename(ENV_NAME).split('.')[0])))
config = get_config()
if args.load:
config.session_init = get_model_loader(args.load)
launch_train_with_config(config, SimpleTrainer())
NUM_ACTIONS = init_player.action_space.n
num_files = init_player.files.num_files
if args.task != 'train':
assert args.load is not None
pred = OfflinePredictor(
PredictConfig(model=Model(),
session_init=get_model_loader(args.load),
input_names=['state'],
output_names=['Qvalue']))
# demo pretrained model one episode at a time
if args.task == 'play':
play_n_episodes(
get_player(directory=data_dir,
files_list=test_list,
viz=0.01,
saveGif=args.saveGif,
saveVideo=args.saveVideo,
task='play'), pred, num_files)
# run episodes in parallel and evaluate pretrained model
elif args.task == 'eval':
play_n_episodes(
get_player(directory=data_dir,
files_list=eval_list,
viz=0.01,
saveGif=args.saveGif,
saveVideo=args.saveVideo,
task='eval'), pred, num_files)
else: # train model
logger.set_logger_dir(logger_dir)
config = get_config()
