Gym安装、调用以及注册

0 gym简介

gym是有openai研发，用于开发和比较强化学习算法的工具组件。

其提供了很多可以直接使用的标准强化学习环境。

1 安装

pip install gym
或
git clone https://github.com/openai/gym
cd gym
pip install -e .

2 调用

import gym
env = gym.make('CartPole-v0') # 调用环境
env.reset() # 对环境进行复位，回到初始状态
for_in range(1000): # 开启一个控制循环
   env.render() # env显示
   env.step(env.action_space.sample())# env take a random action，action from action_space当中随机采样的值
env.close()# 结束控制过程，关闭env

3 注册

虽然gym中有丰富的已经写好的环境，但我们不可能仅仅局限于这些环境当中。我们也可能遇到自己的问题，需要自己写一个环境，但也要基于一个这样的基本框架来运行。如何把自己的环境移入gym当中然后进行调用，就成为一个很关键的问题。这个问题可以称之为“注册”。

（1）找到gym的安装路径。可以在python目录下直接搜索site-packages文件夹，其中的gym文件夹就是gym安装目录
（2）使用提供的core.py替换gym中的同名文件(提供的core.py比原始文件多了些检验部分，这些检验部分主要用于检验创建的环境是否满足gym的要求)
（3）进入env/classic_control，添加grid_game.py文件。classic_control是根据grid_game内容所选择的分类，也可以选择其他的。
（4）修改env/classic_control/_init_.py文件。添加from gym.envs.classic_control.grid_game import GridEnv
（5）修改envs/_init_.py文件
    添加register(
                id='GridWorld-v0',
                entry_point='gym.envs,classic_control:GridEnv',
                max_episode_steps=200,
                reward_threshold=100.0,
    )

core.py

from gym import logger
import numpy as np

import gym
from gym import error
from gym.utils import closer

env_closer = closer.Closer()

# Env-related abstractions

class Env(object):
    """The main OpenAI Gym class. It encapsulates an environment with arbitrary behind-the-scenes dynamics. An environment can be partially or fully observed. The main API methods that users of this class need to know are: transform step reset render close seed And set the following attributes: action_space: The Space object corresponding to valid actions observation_space: The Space object corresponding to valid observations reward_range: A tuple corresponding to the min and max possible rewards Note: a default reward range set to [-inf,+inf] already exists. Set it if you want a narrower range. The methods are accessed publicly as "step", "reset", etc.. The non-underscored versions are wrapper methods to which we may add functionality over time. """

    # Set this in SOME subclasses
    metadata = {
    'render.modes': []}
    reward_range = (-np.inf, np.inf)
    spec = None

    # Set these in ALL subclasses
    action_space = None
    observation_space = None

    def transform(self,state,action):
        ''' Run one timestep of the environment's dynamics. When end of episode is reached, you are responsible for calling `transform()` to reset args state. Accepts an state and action returns a tuple(observation,reward,done,info). Args: state (object):an state provided by the anvironment Returns: Returns: observation (object): agent's observation of the current environment reward (float) : amount of reward returned after previous action done (boolean): whether the episode has ended, in which case further step() calls will return undefined results info (dict): contains auxiliary diagnostic information (helpful for debugging, and sometimes learning) '''
        pass

    def step(self, action):
        """Run one timestep of the environment's dynamics. When end of episode is reached, you are responsible for calling `reset()` to reset this environment's state. Accepts an action and returns a tuple (observation, reward, done, info). Args: action (object): an action provided by the environment Returns: observation (object): agent's observation of the current environment reward (float) : amount of reward returned after previous action done (boolean): whether the episode has ended, in which case further step() calls will return undefined results info (dict): contains auxiliary diagnostic information (helpful for debugging, and sometimes learning) """
        raise NotImplementedError

    def reset(self):
        """Resets the state of the environment and returns an initial observation. Returns: observation (object): the initial observation of the space. """
        raise NotImplementedError

    def render(self, mode='human'):
        """Renders the environment. The set of supported modes varies per environment. (And some environments do not support rendering at all.) By convention, if mode is: - human: render to the current display or terminal and return nothing. Usually for human consumption. - rgb_array: Return an numpy.ndarray with shape (x, y, 3), representing RGB values for an x-by-y pixel image, suitable for turning into a video. - ansi: Return a string (str) or StringIO.StringIO containing a terminal-style text representation. The text can include newlines and ANSI escape sequences (e.g. for colors). Note: Make sure that your class's metadata 'render.modes' key includes the list of supported modes. It's recommended to call super() in implementations to use the functionality of this method. Args: mode (str): the mode to render with close (bool): close all open renderings Example: class MyEnv(Env): metadata = {'render.modes': ['human', 'rgb_array']} def render(self, mode='human'): if mode == 'rgb_array': return np.array(...) # return RGB frame suitable for video elif mode is 'human': ... # pop up a window and render else: super(MyEnv, self).render(mode=mode) # just raise an exception """
        raise NotImplementedError

    def close(self):
        """Override _close in your subclass to perform any necessary cleanup. Environments will automatically close() themselves when garbage collected or when the program exits. """
        return

    def seed(self, seed=None):
        """Sets the seed for this env's random number generator(s). Note: Some environments use multiple pseudorandom number generators. We want to capture all such seeds used in order to ensure that there aren't accidental correlations between multiple generators. Returns: list<bigint>: Returns the list of seeds used in this env's random number generators. The first value in the list should be the "main" seed, or the value which a reproducer should pass to 'seed'. Often, the main seed equals the provided 'seed', but this won't be true if seed=None, for example. """
        logger.warn("Could not seed environment %s", self)
        return

    @property
    def unwrapped(self):
        """Completely unwrap this env. Returns: gym.Env: The base non-wrapped gym.Env instance """
        return self

    def __str__(self):
        if self.spec is None:
            return '<{} instance>'.format(type(self).__name__)
        else:
            return '<{}<{}>>'.format(type(self).__name__, self.spec.id)


class GoalEnv(Env):
    """A goal-based environment. It functions just as any regular OpenAI Gym environment but it imposes a required structure on the observation_space. More concretely, the observation space is required to contain at least three elements, namely `observation`, `desired_goal`, and `achieved_goal`. Here, `desired_goal` specifies the goal that the agent should attempt to achieve. `achieved_goal` is the goal that it currently achieved instead. `observation` contains the actual observations of the environment as per usual. """

    def reset(self):
        # Enforce that each GoalEnv uses a Goal-compatible observation space.
        if not isinstance(self.observation_space, gym.spaces.Dict):
            raise error.Error('GoalEnv requires an observation space of type gym.spaces.Dict')
        result = super(GoalEnv, self).reset()
        for key in ['observation', 'achieved_goal', 'desired_goal']:
            if key not in result:
                raise error.Error('GoalEnv requires the "{}" key to be part of the observation dictionary.'.format(key))
        return result

    def compute_reward(self, achieved_goal, desired_goal, info):
        """Compute the step reward. This externalizes the reward function and makes it dependent on an a desired goal and the one that was achieved. If you wish to include additional rewards that are independent of the goal, you can include the necessary values to derive it in info and compute it accordingly. Args: achieved_goal (object): the goal that was achieved during execution desired_goal (object): the desired goal that we asked the agent to attempt to achieve info (dict): an info dictionary with additional information Returns: float: The reward that corresponds to the provided achieved goal w.r.t. to the desired goal. Note that the following should always hold true: ob, reward, done, info = env.step() assert reward == env.compute_reward(ob['achieved_goal'], ob['goal'], info) """
        raise NotImplementedError()

# Space-related abstractions

class Space(object):
    """Defines the observation and action spaces, so you can write generic code that applies to any Env. For example, you can choose a random action. """
    def __init__(self, shape=None, dtype=None):
        self.shape = None if shape is None else tuple(shape)
        self.dtype = None if dtype is None else np.dtype(dtype)

    def sample(self):
        """ Uniformly randomly sample a random element of this space """
        raise NotImplementedError

    def contains(self, x):
        """ Return boolean specifying if x is a valid member of this space """
        raise NotImplementedError

    def to_jsonable(self, sample_n):
        """Convert a batch of samples from this space to a JSONable data type."""
        # By default, assume identity is JSONable
        return sample_n

    def from_jsonable(self, sample_n):
        """Convert a JSONable data type to a batch of samples from this space."""
        # By default, assume identity is JSONable
        return sample_n


warn_once = True

def deprecated_warn_once(text):
    global warn_once
    if not warn_once: return
    warn_once = False
    logger.warn(text)


class Wrapper(Env):
    env = None

    def __init__(self, env):
        self.env = env
        self.action_space = self.env.action_space
        self.observation_space = self.env.observation_space
        self.reward_range = self.env.reward_range
        self.metadata = self.env.metadata
        self._warn_double_wrap()

    @classmethod
    def class_name(cls):
        return cls.__name__

    def _warn_double_wrap(self):
        env = self.env
        while True:
            if isinstance(env, Wrapper):
                if env.class_name() == self.class_name():
                    raise error.DoubleWrapperError("Attempted to double wrap with Wrapper: {}".format(self.__class__.__name__))
                env = env.env
            else:
                break

    def step(self, action):
        if hasattr(self, "_step"):
            deprecated_warn_once("%s doesn't implement 'step' method, but it implements deprecated '_step' method." % type(self))
            self.step = self._step
            return self.step(action)
        else:
            deprecated_warn_once("%s doesn't implement 'step' method, " % type(self) +
                "which is required for wrappers derived directly from Wrapper. Deprecated default implementation is used.")
            return self.env.step(action)

    def reset(self, **kwargs):
        if hasattr(self, "_reset"):
            deprecated_warn_once("%s doesn't implement 'reset' method, but it implements deprecated '_reset' method." % type(self))
            self.reset = self._reset
            return self._reset(**kwargs)
        else:
            deprecated_warn_once("%s doesn't implement 'reset' method, " % type(self) +
                "which is required for wrappers derived directly from Wrapper. Deprecated default implementation is used.")
            return self.env.reset(**kwargs)

    def render(self, mode='human'):
        return self.env.render(mode)

    def close(self):
        if self.env:
            return self.env.close()

    def seed(self, seed=None):
        return self.env.seed(seed)

    def compute_reward(self, achieved_goal, desired_goal, info):
        return self.env.compute_reward(achieved_goal, desired_goal, info)

    def __str__(self):
        return '<{}{}>'.format(type(self).__name__, self.env)

    def __repr__(self):
        return str(self)

    @property
    def unwrapped(self):
        return self.env.unwrapped

    @property
    def spec(self):
        return self.env.spec


class ObservationWrapper(Wrapper):
    def step(self, action):
        observation, reward, done, info = self.env.step(action)
        return self.observation(observation), reward, done, info

    def reset(self, **kwargs):
        observation = self.env.reset(**kwargs)
        return self.observation(observation)

    def observation(self, observation):
        deprecated_warn_once("%s doesn't implement 'observation' method. Maybe it implements deprecated '_observation' method." % type(self))
        return self._observation(observation)


class RewardWrapper(Wrapper):
    def reset(self):
        return self.env.reset()

    def step(self, action):
        observation, reward, done, info = self.env.step(action)
        return observation, self.reward(reward), done, info

    def reward(self, reward):
        deprecated_warn_once("%s doesn't implement 'reward' method. Maybe it implements deprecated '_reward' method." % type(self))
        return self._reward(reward)


class ActionWrapper(Wrapper):
    def step(self, action):
        action = self.action(action)
        return self.env.step(action)

    def reset(self):
        return self.env.reset()

    def action(self, action):
        deprecated_warn_once("%s doesn't implement 'action' method. Maybe it implements deprecated '_action' method." % type(self))
        return self._action(action)

    def reverse_action(self, action):
        deprecated_warn_once("%s doesn't implement 'reverse_action' method. Maybe it implements deprecated '_reverse_action' method." % type(self))
        return self._reverse_action(action)

grid_game.py

import logging
import numpy
import random
import gym
from gym import spaces

logger = logging.getLogger(__name__)


class GridEnv(gym.Env):
    metadata = {
    
        'render.modes': ['human', 'rgb_array'],
        'video.frames_per_second': 2
    }

    def __init__(self):

        self.states = [0, 1, 2, 3, 4, 5, 6, 7]
        self.x = [140, 220, 300, 380, 460, 140, 300, 460]
        self.y = [250, 250, 250, 250, 250, 150, 150, 150]
        self.terminate_states = dict()  # 终止状态
        self.terminate_states[5] = 1
        self.terminate_states[6] = 1
        self.terminate_states[7] = 1

        self.actions = ['n', 'e', 's', 'w']

        self.rewards = dict()  # 回报的数据结构为字典
        self.rewards['0_s'] = -1.0
        self.rewards['2_s'] = 1.0
        self.rewards['4_s'] = -1.0

        self.t = dict()  # 状态转移的数据格式为字典
        self.t['0_e'] = 1
        self.t['0_s'] = 5
        self.t['1_e'] = 2
        self.t['1_w'] = 0
        self.t['2_e'] = 3
        self.t['2_s'] = 6
        self.t['2_w'] = 1
        self.t['3_e'] = 4
        self.t['3_w'] = 2
        self.t['4_s'] = 7
        self.t['4_w'] = 3

        self.action_space = spaces.Discrete(4)
        self.observation_space = spaces.Discrete(8)

        self.gamma = 0.8  # 折扣因子
        self.viewer = None
        self.state = None

    def getTerminal(self):
        return self.terminate_states

    def getGamma(self):
        return self.gamma

    def getStates(self):
        return self.states

    def getAction(self):
        return self.actions

    def getTerminate_states(self):
        return self.terminate_states

    def setAction(self, s):
        self.state = s

    def _seed(self, seed=None):
        self.np_random, seed = seeding.np_random(seed)
        return seed

    def _step(self, action):
        # 系统当前状态
        state = self.state
        if state in self.terminate_states:
            return state, 0, True, {
    }
        key = "%d_%s" % (state, self.actions[action])  # 将状态和动作组成字典的键值

        # 状态转移
        if key in self.t:
            next_state = self.t[key]
        else:
            next_state = state
        self.state = next_state

        is_terminal = False

        if next_state in self.terminate_states:
            is_terminal = True

        if key not in self.rewards:
            r = 0.0
        else:
            r = self.rewards[key]

        return next_state, r, is_terminal, {
    }

    def _reset(self):
        self.state = self.states[int(random.random() * len(self.states))]
        return self.state

    def _render(self, mode='human', close=False):
        if close:
            if self.viewer is not None:
                self.viewer.close()
                self.viewer = None
            return
        screen_width = 600
        screen_height = 400

        if self.viewer is None:
            from gym.envs.classic_control import rendering
            self.viewer = rendering.Viewer(screen_width, screen_height)
            # 创建网格世界
            self.line1 = rendering.Line((100, 300), (500, 300))
            self.line2 = rendering.Line((100, 200), (500, 200))
            self.line3 = rendering.Line((100, 300), (100, 100))
            self.line4 = rendering.Line((180, 300), (180, 100))
            self.line5 = rendering.Line((260, 300), (260, 100))
            self.line6 = rendering.Line((340, 300), (340, 100))
            self.line7 = rendering.Line((420, 300), (420, 100))
            self.line8 = rendering.Line((500, 300), (500, 100))
            self.line9 = rendering.Line((100, 100), (180, 100))
            self.line10 = rendering.Line((260, 100), (340, 100))
            self.line11 = rendering.Line((420, 100), (500, 100))
            # 创建第一个骷髅
            self.kulo1 = rendering.make_circle(40)
            self.circletrans = rendering.Transform(translation=(140, 150))
            self.kulo1.add_attr(self.circletrans)
            self.kulo1.set_color(0, 0, 0)
            # 创建第二个骷髅
            self.kulo2 = rendering.make_circle(40)
            self.circletrans = rendering.Transform(translation=(460, 150))
            self.kulo2.add_attr(self.circletrans)
            self.kulo2.set_color(0, 0, 0)
            # 创建金条
            self.gold = rendering.make_circle(40)
            self.circletrans = rendering.Transform(translation=(300, 150))
            self.gold.add_attr(self.circletrans)
            self.gold.set_color(1, 0.9, 0)
            # 创建机器人
            self.robot = rendering.make_circle(30)
            self.robotrans = rendering.Transform()
            self.robot.add_attr(self.robotrans)
            self.robot.set_color(0.8, 0.6, 0.4)

            self.line1.set_color(0, 0, 0)
            self.line2.set_color(0, 0, 0)
            self.line3.set_color(0, 0, 0)
            self.line4.set_color(0, 0, 0)
            self.line5.set_color(0, 0, 0)
            self.line6.set_color(0, 0, 0)
            self.line7.set_color(0, 0, 0)
            self.line8.set_color(0, 0, 0)
            self.line9.set_color(0, 0, 0)
            self.line10.set_color(0, 0, 0)
            self.line11.set_color(0, 0, 0)

            self.viewer.add_geom(self.line1)
            self.viewer.add_geom(self.line2)
            self.viewer.add_geom(self.line3)
            self.viewer.add_geom(self.line4)
            self.viewer.add_geom(self.line5)
            self.viewer.add_geom(self.line6)
            self.viewer.add_geom(self.line7)
            self.viewer.add_geom(self.line8)
            self.viewer.add_geom(self.line9)
            self.viewer.add_geom(self.line10)
            self.viewer.add_geom(self.line11)
            self.viewer.add_geom(self.kulo1)
            self.viewer.add_geom(self.kulo2)
            self.viewer.add_geom(self.gold)
            self.viewer.add_geom(self.robot)

        if self.state is None: return None
        # self.robotrans.set_translation(self.x[self.state-1],self.y[self.state-1])
        self.robotrans.set_translation(self.x[self.state], self.y[self.state])

        return self.viewer.render(return_rgb_array=mode == 'rgb_array')