def train_dqn(): args = DQNArgs() env = gym.make(args.env_name) agent = DQNAgent(env, QNet, SimpleNormalizer, args) pre_best = -1e9 for ep in range(args.max_ep): agent.train_one_episode() if ep % args.test_interval == 0: r = agent.test_model() if r > pre_best: pre_best = r agent.save(args.save_dir)