前言

在深度强化学习的实验中，Atari 游戏占了很大的地位。现在我们一般使用 OpenAI 开发的 Gym 包来进行与环境的交互。本文介绍在 Atari 游戏的一些常见预处理过程。

该文所涉及到的 wrapper 均来自 OpenAI baselines
https://github.com/openai/gym…

一些常见 Wrapper 解读

Noop Reset

class NoopResetEnv(gym.Wrapper):
    def __init__(self, env, noop_max=30):
        """Sample initial states by taking random number of no-ops on reset.
        No-op is assumed to be action 0.
        """
        gym.Wrapper.__init__(self, env)
        self.noop_max = noop_max
        self.override_num_noops = None
        self.noop_action = 0
        assert env.unwrapped.get_action_meanings()[0] == 'NOOP'
 
    def reset(self, **kwargs):
        """Do no-op action for a number of steps in [1, noop_max]."""
        self.env.reset(**kwargs)
        if self.override_num_noops is not None:
            noops = self.override_num_noops
        else:
            noops = self.unwrapped.np_random.randint(1, self.noop_max + 1)
        assert noops > 0
        obs = None
        for _ in range(noops):
            obs, _, done, _ = self.env.step(self.noop_action)
            if done:
                obs = self.env.reset(**kwargs)
        return obs
 
    def step(self, ac):
        return self.env.step(ac)

该 wrapper 的作用是在 reset 环境的时候，使用随机数量的 no-op 动作（假设其为环境的动作 0）来采样初始化状态，如果在中途环境已经返回 done 了，则重新 reset 环境。这有利于增加初始画面的随机性，减小陷入过拟合的几率。

Fire Reset

class FireResetEnv(gym.Wrapper):
    def __init__(self, env):
        """Take action on reset for environments that are fixed until firing."""
        gym.Wrapper.__init__(self, env)
        assert env.unwrapped.get_action_meanings()[1] == 'FIRE'
        assert len(env.unwrapped.get_action_meanings()) >= 3
 
    def reset(self, **kwargs):
        self.env.reset(**kwargs)
        obs, _, done, _ = self.env.step(1)
        if done:
            self.env.reset(**kwargs)
        obs, _, done, _ = self.env.step(2)
        if done:
            self.env.reset(**kwargs)
        return obs
 
    def step(self, ac):
        return self.env.step(ac)

在一些 Atari 游戏中，有开火键，比如 Space Invaders，该 wrapper 的作用是返回一个选择开火动作后不 done 的环境状态。

Episodic Life

class EpisodicLifeEnv(gym.Wrapper):
    def __init__(self, env):
        """Make end-of-life == end-of-episode, but only reset on true game over.
        Done by DeepMind for the DQN and co. since it helps value estimation.
        """
        gym.Wrapper.__init__(self, env)
        self.lives = 0
        self.was_real_done  = True
 
    def step(self, action):
        obs, reward, done, info = self.env.step(action)
        self.was_real_done = done
        # check current lives, make loss of life terminal,
        # then update lives to handle bonus lives
        lives = self.env.unwrapped.ale.lives()
        if lives < self.lives and lives > 0:
            # for Qbert sometimes we stay in lives == 0 condition for a few frames
            # so it's important to keep lives > 0, so that we only reset once
            # the environment advertises done.
            done = True
        self.lives = lives
        return obs, reward, done, info
 
    def reset(self, **kwargs):
        """Reset only when lives are exhausted.
        This way all states are still reachable even though lives are episodic,
        and the learner need not know about any of this behind-the-scenes.
        """
        if self.was_real_done:
            obs = self.env.reset(**kwargs)
        else:
            # no-op step to advance from terminal/lost life state
            obs, _, _, _ = self.env.step(0)
        self.lives = self.env.unwrapped.ale.lives()
        return obs

在许多游戏中，玩家操纵的角色有不止一条命，为了加速 Agent 的训练，使其尽量避免死亡，将每条命死亡后的 done 设为 True，同时使用一个属性self.was_real_done 来标记所有生命都用完之后的真正 done。

Max adn Skip

class MaxAndSkipEnv(gym.Wrapper):
    def __init__(self, env, skip=4):
        """Return only every `skip`-th frame"""
        gym.Wrapper.__init__(self, env)
        # most recent raw observations (for max pooling across time steps)
        self._obs_buffer = np.zeros((2,)+env.observation_space.shape, dtype=np.uint8)
        self._skip       = skip
 
    def step(self, action):
        """Repeat action, sum reward, and max over last observations."""
        total_reward = 0.0
        done = None
        for i in range(self._skip):
            obs, reward, done, info = self.env.step(action)
            if i == self._skip - 2: self._obs_buffer[0] = obs
            if i == self._skip - 1: self._obs_buffer[1] = obs
            total_reward += reward
            if done:
                break
        # Note that the observation on the done=True frame
        # doesn't matter
        max_frame = self._obs_buffer.max(axis=0)
 
        return max_frame, total_reward, done, info
 
    def reset(self, **kwargs):
        return self.env.reset(**kwargs)

该 Wrapper 提供了跳帧操作，即每 skip 帧返回一次环境状态元组，在跳过的帧里执行相同的动作，将其奖励叠加，并且取最后两帧像素值中的最大值。在 Atari 游戏中，有些画面是仅在奇数帧出现的，因此要对最后两帧取最大值。

Clip Reward

class ClipRewardEnv(gym.RewardWrapper):
    def __init__(self, env):
        gym.RewardWrapper.__init__(self, env)
 
    def reward(self, reward):
        """Bin reward to {+1, 0, -1} by its sign."""
        return np.sign(reward)

对于不同游戏来说，其得分衡量也是不同的，为了便于统一度量和学习，将所有奖励统一定义为 1（reward > 0），0（reward = 0）或 -1（reward < 0）。

Wrap Frame

class WarpFrame(gym.ObservationWrapper):
    def __init__(self, env, width=84, height=84, grayscale=True, dict_space_key=None):
        """
        Warp frames to 84x84 as done in the Nature paper and later work.
 
        If the environment uses dictionary observations, `dict_space_key` can be specified which indicates which
        observation should be warped.
        """
        super().__init__(env)
        self._width = width
        self._height = height
        self._grayscale = grayscale
        self._key = dict_space_key
        if self._grayscale:
            num_colors = 1
        else:
            num_colors = 3
 
        new_space = gym.spaces.Box(
            low=0,
            high=255,
            shape=(self._height, self._width, num_colors),
            dtype=np.uint8,
        )
        if self._key is None:
            original_space = self.observation_space
            self.observation_space = new_space
        else:
            original_space = self.observation_space.spaces[self._key]
            self.observation_space.spaces[self._key] = new_space
        assert original_space.dtype == np.uint8 and len(original_space.shape) == 3
 
    def observation(self, obs):
        if self._key is None:
            frame = obs
        else:
            frame = obs[self._key]
 
        if self._grayscale:
            frame = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)
        frame = cv2.resize(frame[34:194], (self._width, self._height), interpolation=cv2.INTER_AREA
        )
        if self._grayscale:
            frame = np.expand_dims(frame, -1)
 
        if self._key is None:
            obs = frame
        else:
            obs = obs.copy()
            obs[self._key] = frame
        return obs

该 Wrap 对观察到的帧的图片数据进行了处理。首先将 3 维 RGB 图像转为灰度图像，之后将其 resize 为 84 × 84 的灰度图像。本例为 Pong 游戏的 wrap Frame 处理，为使 Agent 更关注于游戏本身的画面，避免被得分等图像区域误导，我对画面进行了裁切（frame[34: 194]），对于不同的游戏，裁切的方法可能不同。

Frame Stack

class FrameStack(gym.Wrapper):
    def __init__(self, env, k):
        """Stack k last frames.
 
        Returns lazy array, which is much more memory efficient.
 
        See Also
        --------
        baselines.common.atari_wrappers.LazyFrames
        """
        gym.Wrapper.__init__(self, env)
        self.k = k
        self.frames = deque([], maxlen=k)
        shp = env.observation_space.shape
        self.observation_space = spaces.Box(low=0, high=255, shape=(shp[:-1] + (shp[-1] * k,)), dtype=env.observation_space.dtype)
 
    def reset(self):
        ob = self.env.reset()
        for _ in range(self.k):
            self.frames.append(ob)
        return self._get_ob()
 
    def step(self, action):
        ob, reward, done, info = self.env.step(action)
        self.frames.append(ob)
        return self._get_ob(), reward, done, info
 
    def _get_ob(self):
        assert len(self.frames) == self.k
        return LazyFrames(list(self.frames))

本 wrapper 的作用是将 k 帧灰度图像并为一帧，以此来为 CNN 提供一些序列信息（Human Level control through deep reinforcement learning）。Wrapper 会维持一个大小为 k 的 deque，之后依次使用最新的 ob 来替代最久远的 ob，达到不同时间的状态叠加的效果。最后返回一个 LazyFrame。如果想要使用 LazyFrame，只需利用 np.array(lazy_frames_instance) 即可将 LazyFrame 对象转为 ndarray 对象。

Scaled Float Frame

class ScaledFloatFrame(gym.ObservationWrapper):
    def __init__(self, env):
        gym.ObservationWrapper.__init__(self, env)
        self.observation_space = gym.spaces.Box(low=0, high=1, shape=env.observation_space.shape, dtype=np.float32)
 
    def observation(self, observation):
        # careful! This undoes the memory optimization, use
        # with smaller replay buffers only.
        return (np.array(observation).astype(np.float32) / 255.0

该 Wrapper 的目的是将 0 ~ 255 的图像归一化到 [0, 1]。

后记

在强化学习训练中，环境及其预处理是一个非常重要的步骤，甚至可以直接影响到强化学习智能体的训练成功与否。除上述 Wrapper 外，读者也可另根据自己的需求来写 Wrapper，以满足需求。最后，附上 Nature DQN 的环境 Wrapper：

def wrap_deepmind(env, episode_life=True, clip_rewards=True, frame_stack=False, scale=False):
    """Configure environment for DeepMind-style Atari."""
    if episode_life:
        env = EpisodicLifeEnv(env)
    if 'FIRE' in env.unwrapped.get_action_meanings():
        env = FireResetEnv(env)
    env = WarpFrame(env)
    if scale:
        env = ScaledFloatFrame(env)
    if clip_rewards:
        env = ClipRewardEnv(env)
    if frame_stack:
        env = FrameStack(env, 4)
    return env

PS

找了一圈博客网站，CSDN 太恶心直接拉黑，博客园又要求实名信息，总感觉不爽，最后选择了 SegmentFault。其实最终还是希望能使用自建的博客的，不过一来暂时没有主机，二来最近也没时间折腾 GitHub Pages，先在这里记录一下学习笔记好了。就酱。