一些深度强化学习相关的开源库和学习资料

今天药忘吃喽~ 2022-08-29 00:09 303阅读 0赞

为了回答深度强化学习实验室论坛的问题，按照自己的github梳理了关注的一些开源库：

深度强化学习实验室自己的官方资料汇总链接，感觉还是很好用的：  
[https://github.com/NeuronDance/DeepRL][https_github.com_NeuronDance_DeepRL]

基于torch的单强化学习开源库，里面用面向对象的方法封装了几乎所有主流的单智能体强化学习pytorch实现：  
[https://github.com/p-christ/Deep-Reinforcement-Learning-Algorithms-with-PyTorch][https_github.com_p-christ_Deep-Reinforcement-Learning-Algorithms-with-PyTorch]

基于torch的多智能体强化学习开源库，用星际争霸案例对State of the art的MARL算法进行了实现：  
[https://github.com/starry-sky6688/StarCraft][https_github.com_starry-sky6688_StarCraft]

分布式强化学习开源库ray：  
[https://github.com/ray-project/ray][https_github.com_ray-project_ray]

基于torch的另外几个单强化学习开源库，个人觉得都没有上面第二个推荐那个torch的好用，但是也为一些算法的复现提供了不错的思路：  
[https://github.com/p-christ/Deep-Reinforcement-Learning-Algorithms-with-PyTorch][https_github.com_p-christ_Deep-Reinforcement-Learning-Algorithms-with-PyTorch]  
[https://github.com/sweetice/Deep-reinforcement-learning-with-pytorch][https_github.com_sweetice_Deep-reinforcement-learning-with-pytorch]  
[https://github.com/astooke/rlpyt][https_github.com_astooke_rlpyt]

一个国内团队开发的强化学习算法开发框架库ElegantRl，主打轻量化、稳定、高效开发：  
[https://github.com/AI4Finance-LLC/ElegantRL][https_github.com_AI4Finance-LLC_ElegantRL]  
(附上知乎的介绍链接：[https://zhuanlan.zhihu.com/p/127792558][https_zhuanlan.zhihu.com_p_127792558])

OpenAI开发的一个多智能体混合合作竞争的环境库，当时为了验证MADDPG算法，后来也用于其他多智能体强化算法的验证：  
[https://github.com/openai/multiagent-particle-envs][https_github.com_openai_multiagent-particle-envs]  
（附上对于环境的一个中文博客介绍：[https://blog.csdn.net/weixin\_43483381/article/details/114240352?spm=1001.2014.3001.5501][https_blog.csdn.net_weixin_43483381_article_details_114240352_spm_1001.2014.3001.5501]）

多智能体强化学习竞争合作关系下的算法开创，MADDPG，openAI组织的，开源代码：  
[https://github.com/starry-sky6688/MADDPG][https_github.com_starry-sky6688_MADDPG]

一个初级的多智能体强化学习算法框架库，不如上边推荐那个好用，但是可以参考：  
[https://github.com/oxwhirl/pymarl][https_github.com_oxwhirl_pymarl]

多智能体强化学习算法的一个可视化环境，也是OpenAI那个强化学习seek and hide视频用的环境:  
[https://github.com/openai/multi-agent-emergence-environments][https_github.com_openai_multi-agent-emergence-environments]  
(附上seek and hide视频链接，用的MADDPG控制：[https://www.bilibili.com/video/BV17e411x7sh][https_www.bilibili.com_video_BV17e411x7sh])

一位本科上交研究生出国的学长的总结的MARL相关的论文，当时看了受益很大：  
[https://github.com/LantaoYu/MARL-Paper][https_github.com_LantaoYu_MARL-Paper]

教学视频的话几个推荐：  
1）国产李宏毅老师的强化学习教学，里面对Policy-based的讲解尤其清晰：  
[https://www.bilibili.com/video/BV124411S7au][https_www.bilibili.com_video_BV124411S7au]

2）斯坦福的CS234课程，当时博一上课的时候老师推荐的，里面对于RL怎么来的讲的尤其清晰：  
[https://www.bilibili.com/video/BV1Cc411h7QQ][https_www.bilibili.com_video_BV1Cc411h7QQ]

3）之前从外网看到的wangshusen老师的视频，感觉也很清晰，其课程ppt和视频都在下面的链接里：  
[https://github.com/wangshusen/DRL][https_github.com_wangshusen_DRL]

4）莫烦的python强化学习视频，视频做的很生动有趣：  
[https://www.bilibili.com/video/BV13W411Y75P?from=search&seid=14400823287014350597][https_www.bilibili.com_video_BV13W411Y75P_from_search_seid_14400823287014350597]

5）最后推荐一个小的但是很有意思的视频，讲神经网络反向传播的，生动形象：  
[https://www.bilibili.com/video/BV16x411V7Qg][https_www.bilibili.com_video_BV16x411V7Qg]

[https_github.com_NeuronDance_DeepRL]: https://github.com/NeuronDance/DeepRL
[https_github.com_p-christ_Deep-Reinforcement-Learning-Algorithms-with-PyTorch]: https://github.com/p-christ/Deep-Reinforcement-Learning-Algorithms-with-PyTorch
[https_github.com_starry-sky6688_StarCraft]: https://github.com/starry-sky6688/StarCraft
[https_github.com_ray-project_ray]: https://github.com/ray-project/ray
[https_github.com_sweetice_Deep-reinforcement-learning-with-pytorch]: https://github.com/sweetice/Deep-reinforcement-learning-with-pytorch
[https_github.com_astooke_rlpyt]: https://github.com/astooke/rlpyt
[https_github.com_AI4Finance-LLC_ElegantRL]: https://github.com/AI4Finance-LLC/ElegantRL
[https_zhuanlan.zhihu.com_p_127792558]: https://zhuanlan.zhihu.com/p/127792558
[https_github.com_openai_multiagent-particle-envs]: https://github.com/openai/multiagent-particle-envs
[https_blog.csdn.net_weixin_43483381_article_details_114240352_spm_1001.2014.3001.5501]: https://blog.csdn.net/weixin_43483381/article/details/114240352?spm=1001.2014.3001.5501
[https_github.com_starry-sky6688_MADDPG]: https://github.com/starry-sky6688/MADDPG
[https_github.com_oxwhirl_pymarl]: https://github.com/oxwhirl/pymarl
[https_github.com_openai_multi-agent-emergence-environments]: https://github.com/openai/multi-agent-emergence-environments
[https_www.bilibili.com_video_BV17e411x7sh]: https://www.bilibili.com/video/BV17e411x7sh
[https_github.com_LantaoYu_MARL-Paper]: https://github.com/LantaoYu/MARL-Paper
[https_www.bilibili.com_video_BV124411S7au]: https://www.bilibili.com/video/BV124411S7au
[https_www.bilibili.com_video_BV1Cc411h7QQ]: https://www.bilibili.com/video/BV1Cc411h7QQ
[https_github.com_wangshusen_DRL]: https://github.com/wangshusen/DRL
[https_www.bilibili.com_video_BV13W411Y75P_from_search_seid_14400823287014350597]: https://www.bilibili.com/video/BV13W411Y75P?from=search&seid=14400823287014350597
[https_www.bilibili.com_video_BV16x411V7Qg]: https://www.bilibili.com/video/BV16x411V7Qg