def run_parallel_episodes(arg):
total_steps = 0
env_c = gym.Init(arg[4], False)
policy = Policy(arg[0], arg[1], arg[2], arg[4], True)
scaler = Scaler(arg[0], arg[4])
scaler.resume()
observes, actions, rewards, unscaled_obs = run_episode(env_c, policy, scaler, arg[3])
total_steps += observes.shape[0]
trajectory = {'observes': observes,
'actions': actions,
'rewards': rewards,
'unscaled_obs': unscaled_obs}
policy.close_sess()
return trajectory
def init_gym(env_name, render):
"""
Initialize gym environment, return dimension of observation
and action spaces.
Args:
render: True to toggle on visualization
Returns: 3-tuple
environment (object)
number of observation dimensions (int)
number of action dimensions (int)
"""
env = gym.Init(env_name, render)
obs_dim = env.observation_space.shape[0]
act_dim = env.action_space.shape[0]
return env, obs_dim, act_dim
