n_steps = 0
os.makedirs(log_dir, exist_ok=True)
################ MODEL AND GYM ENVIRONMENT
if ENVIRONMENT == 'rgbd':
env = tm700_rgbd_gym(renders=RENDERS, isDiscrete=DISCRETE)
env = Monitor(env, os.path.join(log_dir, 'monitor.csv'), allow_early_resets=True)
if ENVIRONMENT == 'rgb':
env = tm700_rgb_gym(renders=RENDERS, isDiscrete=DISCRETE)
env = Monitor(env, os.path.join(log_dir, 'monitor.csv'), allow_early_resets=True)
if ENVIRONMENT == 'possensor':
env = tm700_possensor_gym(renders=RENDERS, isDiscrete=DISCRETE)
env = Monitor(env, os.path.join(log_dir, 'monitor.csv'), allow_early_resets=True)
if MODEL == 'DQN':
from stable_baselines.deepq.policies import LnCnnPolicy, MlpPolicy
if ENVIRONMENT in ['rgbd', 'rgb', 'rgbdsparse']:
model = DQN(LnCnnPolicy, env, verbose=1,
tensorboard_log=(log_dir + "tensorboard_%s_%s_%s/") % (MODEL, ENVIRONMENT, DATE),
gamma=0.99, learning_rate=0.0001, buffer_size=50000, exploration_fraction=0.1, exploration_final_eps=0.02,
train_freq=1, batch_size=32, double_q=True, learning_starts=1000,
target_network_update_freq=500, prioritized_replay=True, prioritized_replay_alpha=0.6,
prioritized_replay_beta0=0.4, prioritized_replay_beta_iters=None, prioritized_replay_eps=1e-06,
param_noise=False, _init_setup_model=True,
policy_kwargs=None, full_tensorboard_log=False)
import logging
tf.get_logger().setLevel(logging.ERROR)
args=parser.arg_parse()
# Create save dir
model_dir = args.model_dir
if not os.path.exists(args.model_dir):
os.makedirs(args.model_dir)
############## load environment
env = tm700_possensor_gym(renders=False, isDiscrete=True)
# env = gym.make('CartPole-v1')
# vectorized environments allow to easily multiprocess training
# we demonstrate its usefulness in the next examples
env = DummyVecEnv([lambda: env]) # The algorithms require a vectorized environment to run
############ MODELS
model = PPO2('MlpPolicy', 'Pendulum-v0', verbose=0).learn(8000)
# The model will be saved under PPO2_tutorial.zip
# ddpg_model = DDPG(MlpPolicy, env, verbose=1, param_noise=None, random_exploration=0.1)
kwargs = {'double_q': True, 'prioritized_replay': True, 'policy_kwargs': dict(dueling=True)} #DQN + Prioritized Experience Replay + Double Q-Learning + Dueling
dqn_model = DQN('MlpPolicy', env, verbose=1, **kwargs)
