envs.py 文件源码

def make_env(env_id, seed, rank, log_dir):
    def _thunk():
        env = gym.make(env_id)
        is_atari = hasattr(gym.envs, 'atari') and isinstance(env.unwrapped, gym.envs.atari.atari_env.AtariEnv)
        if is_atari:
            env = make_atari(env_id)
        env.seed(seed + rank)
        if log_dir is not None:
            env = bench.Monitor(env, os.path.join(log_dir, str(rank)))
        if is_atari:
            env = wrap_deepmind(env)
        # If the input has shape (W,H,3), wrap for PyTorch convolutions
        obs_shape = env.observation_space.shape
        if len(obs_shape) == 3 and obs_shape[2] in [1, 3]:
            env = WrapPyTorch(env)
        return env

    return _thunk