东方耀AI技术分享

标题: 通过安装gym与retro搭建强化学习的实验环境 [打印本页]

作者: 东方耀 时间: 2019-8-18 07:59
标题: 通过安装gym与retro搭建强化学习的实验环境

通过安装gym与retro搭建强化学习的实验环境

gym下载：https://gitee.com/dfy888/gym

retro下载：https://gitee.com/dfy888/retro

先下载zip源码包通过源码编译的方式安装

系统：Ubuntu18.04

隔离工具：conda

IDE工具：Pycharm

python版本：3.7

安装步骤如下：
1、准备工作（虚拟环境创建与work目录）

conda create -n py3_gym python=3.7
conda activate py3_gym
python3 get-pip.py
pip --version
pip3 --version
cd & mkdir py3_gym_works
将gym源码包解压到 py3_gym_works目录下

复制代码

2、打开pycharm 创建项目关联虚拟环境与py3_gym_works目录
3、开始安装gym

cd gym
pip3 install -e '.[all]'
会报错：
Building wheel for mujoco-py (setup.py) ... error
ERROR: Command errored out with exit status 1:
原因：无法编译mujoco 其需要依赖其他的包
我们这里暂时不需要用到mujoco 故将
setup.py文件中屏蔽mujoco相关的代码
# 'mujoco': ['mujoco_py>=1.50, <2.1', 'imageio'],
# 'robotics': ['mujoco_py>=1.50, <2.1', 'imageio'],
再次运行：
pip3 install -e '.[all]' 提示成功
pip3 list 可以看到安装的gym

复制代码

4、测试gym是否安装成功，跑下面的代码，显示如下图所示

# -*- coding: utf-8 -*-
__author__ = u'东方耀微信：dfy_88888'
__date__ = '2019/8/18 上午08:43'
__product__ = 'PyCharm'
__filename__ = 'demo_gym'
import gym
env = gym.make('CartPole-v0')
for i_episode in range(10):
# Reset the environment's state. Returns observation.
observation = env.reset()
print('i_episode回合数：', i_episode)
for t in range(100):
# Render one frame of the environment.
env.render() # 更新动画
action = env.action_space.sample()
# Step the environment by one timestep. Returns observation, reward, done, info.
observation, reward, done, info = env.step(action) # 推进一步
if done:
env.reset()
continue

复制代码

5、开始编译安装gym-retro，如下图所示

cd retro
pip3 install cmake
pip3 install -e .
sudo apt-get install capnproto libcapnp-dev libqt5opengl5-dev qtbase5-dev
cmake . -DBUILD_UI=ON -UPYLIB_DIRECTORY
make -j$(grep -c ^processor /proc/cpuinfo)
./gym-retro-integration

复制代码

6、代码测试retro是否成功，如下图所示

# -*- coding: utf-8 -*-
__author__ = u'东方耀微信：dfy_88888'
__date__ = '2019/8/18 上午09:15'
__product__ = 'PyCharm'
__filename__ = 'retro_demo'
import retro
def main():
# Create a Gym Environment
env = retro.make(game='Airstriker-Genesis')
# reset(self): Reset the environment's state. Returns observation.
obs = env.reset()
while True:
# Step the environment by one timestep. Returns observation, reward, done, info.
obs, rew, done, info = env.step(env.action_space.sample())
# render(self, mode='human'): Render one frame of the environment.
# The default mode will do something human friendly, such as pop up a window.
env.render()
if done:
obs = env.reset()
env.close()
if __name__ == "__main__":
main()

复制代码

7、
8、

欢迎光临东方耀AI技术分享 (http://www.ai111.vip/)