def main(_):
logging.set_verbosity(logging.INFO)
gin.parse_config_files_and_bindings(FLAGS.gin_file, FLAGS.gin_bindings)
# Setup log dir.
if FLAGS.sub_dir == 'auto':
sub_dir = utils.get_datetime()
else:
sub_dir = FLAGS.sub_dir
log_dir = os.path.join(
FLAGS.root_dir,
FLAGS.env_name,
FLAGS.identifier,
FLAGS.agent_name,
sub_dir,
str(FLAGS.seed),
)
utils.maybe_makedirs(log_dir)
eval_results = train_eval_offline.train_eval_offline(
log_dir=log_dir,
data_file=None,
agent_module=agents.AGENT_MODULES_DICT[FLAGS.agent_name],
env_name=FLAGS.env_name,
n_train=FLAGS.n_train,
total_train_steps=FLAGS.total_train_steps,
n_eval_episodes=FLAGS.n_eval_episodes,
)
results_file = os.path.join(log_dir, 'results.npy')
with tf.io.gfile.GFile(results_file, 'w') as f:
np.save(f, eval_results)
def main(_):
# Setup log dir.
if FLAGS.sub_dir == 'auto':
sub_dir = utils.get_datetime()
else:
sub_dir = FLAGS.sub_dir
log_dir = os.path.join(
FLAGS.root_dir,
FLAGS.env_name,
'bc',
sub_dir,
str(FLAGS.seed),
)
model_arch = ((200,200),)
opt_params = (('adam', 5e-4),)
utils.maybe_makedirs(log_dir)
train_eval_offline.train_eval_offline(
log_dir=log_dir,
data_file=None,
agent_module=agents.AGENT_MODULES_DICT['bc'],
env_name=FLAGS.env_name,
n_train=FLAGS.n_train,
total_train_steps=FLAGS.total_train_steps,
n_eval_episodes=1,
model_params=model_arch,
optimizers=opt_params,
)
def main(_):
logging.set_verbosity(logging.INFO)
gin.parse_config_files_and_bindings(FLAGS.gin_file, FLAGS.gin_bindings)
# Setup log dir.
if FLAGS.sub_dir == 'auto':
sub_dir = utils.get_datetime()
else:
sub_dir = FLAGS.sub_dir
log_dir = os.path.join(
FLAGS.root_dir,
FLAGS.env_name,
FLAGS.identifier,
FLAGS.agent_name,
sub_dir,
str(FLAGS.seed),
)
utils.maybe_makedirs(log_dir)
model_arch = None
if FLAGS.model_arch == 0:
model_arch = ((200, 200), )
elif FLAGS.model_arch == 1:
model_arch = ((
(300, 300),
(200, 200),
), 2)
else:
raise ValueError()
if FLAGS.opt_params == 0:
opt_params = (('adam', 1e-5), )
elif FLAGS.opt_params == 1:
opt_params = (('adam', 1e-3), ('adam', 3e-5), ('adam', 1e-5))
elif FLAGS.opt_params == 2:
opt_params = (('adam', 1e-3), ('adam', 3e-4), ('adam', 1e-5))
elif FLAGS.opt_params == 3:
opt_params = (('adam', 0e-3), ('adam', 0e-4), ('adam', 0e-5))
else:
raise ValueError()
eval_results = train_eval_offline.train_eval_offline(
log_dir=log_dir,
data_file=None,
agent_module=agents.AGENT_MODULES_DICT[FLAGS.agent_name],
env_name=FLAGS.env_name,
n_train=FLAGS.n_train,
total_train_steps=FLAGS.total_train_steps,
n_eval_episodes=FLAGS.n_eval_episodes,
model_params=model_arch,
optimizers=opt_params,
value_penalty=bool(FLAGS.value_penalty),
behavior_ckpt_file=FLAGS.b_ckpt,
save_freq=FLAGS.save_freq)
results_file = os.path.join(log_dir, 'results.npy')
with tf.io.gfile.GFile(results_file, 'w') as f:
np.save(f, eval_results)
def main(_):
logging.set_verbosity(logging.INFO)
gin.parse_config_files_and_bindings(FLAGS.gin_file, FLAGS.gin_bindings)
sub_dir = FLAGS.sub_dir
log_dir = os.path.join(
FLAGS.root_dir,
FLAGS.env_name,
FLAGS.data_name,
sub_dir,
)
utils.maybe_makedirs(log_dir)
config_module = importlib.import_module('{}.{}'.format(
FLAGS.config_dir, FLAGS.config_file))
collect_data(log_dir=log_dir,
data_config=config_module.get_data_config(
FLAGS.env_name, FLAGS.policy_root_dir),
n_samples=FLAGS.n_samples,
env_name=FLAGS.env_name,
n_eval_episodes=FLAGS.n_eval_episodes)
def main(_):
logging.set_verbosity(logging.INFO)
gin.parse_config_files_and_bindings(FLAGS.gin_file, FLAGS.gin_bindings)
if FLAGS.sub_dir == "auto":
sub_dir = utils.get_datetime()
else:
sub_dir = FLAGS.sub_dir
log_dir = os.path.join(
FLAGS.root_dir,
FLAGS.env_name,
FLAGS.agent_name,
sub_dir,
)
utils.maybe_makedirs(log_dir)
train_eval_online.train_eval_online(
log_dir=log_dir,
agent_module=agents.AGENT_MODULES_DICT[FLAGS.agent_name],
env_name=FLAGS.env_name,
total_train_steps=FLAGS.total_train_steps,
n_eval_episodes=FLAGS.n_eval_episodes,
eval_target=FLAGS.eval_target,
)
def main(_):
logging.set_verbosity(logging.INFO)
gin.parse_config_files_and_bindings(FLAGS.gin_file, FLAGS.gin_bindings)
# Setup data file path.
data_dir = os.path.join(
FLAGS.data_root_dir,
FLAGS.env_name,
FLAGS.data_name,
FLAGS.data_sub_dir,
)
data_file = os.path.join(
data_dir, FLAGS.data_file_name)
# Setup log dir.
if FLAGS.sub_dir == 'auto':
sub_dir = utils.get_datetime()
else:
sub_dir = FLAGS.sub_dir
log_dir = os.path.join(
FLAGS.root_dir,
FLAGS.env_name,
FLAGS.data_name,
'n'+str(FLAGS.n_train),
FLAGS.agent_name,
sub_dir,
str(FLAGS.seed),
)
utils.maybe_makedirs(log_dir)
train_eval_offline.train_eval_offline(
log_dir=log_dir,
data_file=data_file,
agent_module=agents.AGENT_MODULES_DICT[FLAGS.agent_name],
env_name=FLAGS.env_name,
n_train=FLAGS.n_train,
total_train_steps=FLAGS.total_train_steps,
n_eval_episodes=FLAGS.n_eval_episodes,
)
def main(_):
logging.set_verbosity(logging.INFO)
gin.parse_config_files_and_bindings(FLAGS.gin_file, FLAGS.gin_bindings)
bc_log_dir, sub_dir = train_bc()
behavior_ckpt_file = os.path.join(bc_log_dir, 'agent_behavior')
log_dir = os.path.join(
FLAGS.root_dir,
FLAGS.env_name,
FLAGS.agent_name,
sub_dir,
str(FLAGS.seed),
)
model_arch = ((
(300, 300),
(200, 200),
), 2)
opt_params = (('adam', 1e-3), ('adam', 3e-5), ('adam', 1e-5))
utils.maybe_makedirs(log_dir)
train_eval_offline.train_eval_offline(
log_dir=log_dir,
data_file=None,
agent_module=agents.AGENT_MODULES_DICT[FLAGS.agent_name],
env_name=FLAGS.env_name,
n_train=FLAGS.n_train,
total_train_steps=FLAGS.total_train_steps,
n_eval_episodes=FLAGS.n_eval_episodes,
model_params=model_arch,
optimizers=opt_params,
behavior_ckpt_file=behavior_ckpt_file,
value_penalty=bool(FLAGS.value_penalty),
save_freq=FLAGS.save_freq,
alpha=FLAGS.alpha)
