(rank)
| 32 | |
| 33 | def train(env_id, num_timesteps, num_cpu): |
| 34 | def make_env(rank): |
| 35 | def _thunk(): |
| 36 | env = make_atari(env_id) |
| 37 | env.seed(SEED + rank) |
| 38 | gym.logger.setLevel(logging.WARN) |
| 39 | env = wrap_deepmind(env) |
| 40 | |
| 41 | # wrap the env one more time for getting total reward |
| 42 | env = Monitor(env, rank) |
| 43 | return env |
| 44 | return _thunk |
| 45 | |
| 46 | env = SubprocVecEnv([make_env(i) for i in range(num_cpu)]) |
| 47 | learn(CNN, env, SEED, total_timesteps=int(num_timesteps * 1.1)) |