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

Before Change

        epsilon_greedy_selector.epsilon = max(EPSILON_FINAL, EPSILON_START - frame_idx / EPSILON_DECAY_LAST_FRAME)

        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

        optimizer.zero_grad()

After Change

    env = ptan.common.wrappers.wrap_dqn(env)
    env = ptan.common.wrappers.ScaledFloatFrame(env)

    writer = SummaryWriter(comment="-pong-ptan")
    net = dqn_model.DQN(env.observation_space.shape, env.action_space.n)
    if args.cuda:
        net.cuda()

Instances