c4bfe2ab1d355f8ed0c9881fd4f9d114cc627126,ch06/04_dqn_pong_ptan.py,,,#,64

Before Change

        new_rewards = exp_source.pop_total_rewards()
        if new_rewards:
            total_rewards.extend(new_rewards)
            print("%d: %s" % (frame_idx, new_rewards))

        if len(buffer) < REPLAY_START_SIZE:
            continue

After Change

            speed = (frame_idx - ts_frame) / (time.time() - ts)
            ts_frame = frame_idx
            ts = time.time()
            mean_reward = np.mean(total_rewards[-100:])
            print("%d: done %d games, mean reward %.3f, eps %.2f, speed %.2f f/s" % (
                frame_idx, len(total_rewards), mean_reward, epsilon_greedy_selector.epsilon,
                speed
            ))
            writer.add_scalar("epsilon", epsilon_greedy_selector.epsilon, frame_idx)
            writer.add_scalar("speed", speed, frame_idx)
            writer.add_scalar("reward_100", mean_reward, frame_idx)
            writer.add_scalar("reward", new_rewards[0], frame_idx)
            if mean_reward > MEAN_REWARD_BOUND:
                print("Solved in %d frames!" % frame_idx)
                break

Instances