bupticybee / Icychesszero
Licence: mit
中国象棋alpha zero程序
Stars: ✭ 206
Projects that are alternatives of or similar to Icychesszero
Modular Rl
[ICML 2020] PyTorch Code for "One Policy to Control Them All: Shared Modular Policies for Agent-Agnostic Control"
Stars: ✭ 126 (-38.83%)
Mutual labels: jupyter-notebook, reinforcement-learning
Pytorch sac
PyTorch implementation of Soft Actor-Critic (SAC)
Stars: ✭ 174 (-15.53%)
Mutual labels: jupyter-notebook, reinforcement-learning
Data Science Question Answer
A repo for data science related questions and answers
Stars: ✭ 2,000 (+870.87%)
Mutual labels: jupyter-notebook, reinforcement-learning
Advanced Deep Learning And Reinforcement Learning Deepmind
🎮 Advanced Deep Learning and Reinforcement Learning at UCL & DeepMind | YouTube videos 👉
Stars: ✭ 121 (-41.26%)
Mutual labels: jupyter-notebook, reinforcement-learning
Deeprl Agents
A set of Deep Reinforcement Learning Agents implemented in Tensorflow.
Stars: ✭ 2,149 (+943.2%)
Mutual labels: jupyter-notebook, reinforcement-learning
Rl Quadcopter
Teach a Quadcopter How to Fly!
Stars: ✭ 124 (-39.81%)
Mutual labels: jupyter-notebook, reinforcement-learning
2048 Deep Reinforcement Learning
Trained A Convolutional Neural Network To Play 2048 using Deep-Reinforcement Learning
Stars: ✭ 169 (-17.96%)
Mutual labels: jupyter-notebook, reinforcement-learning
Ctc Executioner
Master Thesis: Limit order placement with Reinforcement Learning
Stars: ✭ 112 (-45.63%)
Mutual labels: jupyter-notebook, reinforcement-learning
Andrew Ng Notes
This is Andrew NG Coursera Handwritten Notes.
Stars: ✭ 180 (-12.62%)
Mutual labels: jupyter-notebook, reinforcement-learning
Deep Algotrading
A resource for learning about deep learning techniques from regression to LSTM and Reinforcement Learning using financial data and the fitness functions of algorithmic trading
Stars: ✭ 173 (-16.02%)
Mutual labels: jupyter-notebook, reinforcement-learning
Pytorch Rl
Tutorials for reinforcement learning in PyTorch and Gym by implementing a few of the popular algorithms. [IN PROGRESS]
Stars: ✭ 121 (-41.26%)
Mutual labels: jupyter-notebook, reinforcement-learning
Alpha Zero General
A clean implementation based on AlphaZero for any game in any framework + tutorial + Othello/Gobang/TicTacToe/Connect4 and more
Stars: ✭ 2,617 (+1170.39%)
Mutual labels: jupyter-notebook, reinforcement-learning
Reinforcementlearning Atarigame
Pytorch LSTM RNN for reinforcement learning to play Atari games from OpenAI Universe. We also use Google Deep Mind's Asynchronous Advantage Actor-Critic (A3C) Algorithm. This is much superior and efficient than DQN and obsoletes it. Can play on many games
Stars: ✭ 118 (-42.72%)
Mutual labels: jupyter-notebook, reinforcement-learning
Multihopkg
Multi-hop knowledge graph reasoning learned via policy gradient with reward shaping and action dropout
Stars: ✭ 202 (-1.94%)
Mutual labels: jupyter-notebook, reinforcement-learning
Coursera reinforcement learning
Coursera Reinforcement Learning Specialization by University of Alberta & Alberta Machine Intelligence Institute
Stars: ✭ 114 (-44.66%)
Mutual labels: jupyter-notebook, reinforcement-learning
Chess Alpha Zero
Chess reinforcement learning by AlphaGo Zero methods.
Stars: ✭ 1,868 (+806.8%)
Mutual labels: jupyter-notebook, reinforcement-learning
Tensorflow2.0 Examples
🙄 Difficult algorithm, Simple code.
Stars: ✭ 1,397 (+578.16%)
Mutual labels: jupyter-notebook, reinforcement-learning
Machine Learning And Reinforcement Learning In Finance
Machine Learning and Reinforcement Learning in Finance New York University Tandon School of Engineering
Stars: ✭ 173 (-16.02%)
Mutual labels: jupyter-notebook, reinforcement-learning
Release
Deep Reinforcement Learning for de-novo Drug Design
Stars: ✭ 201 (-2.43%)
Mutual labels: jupyter-notebook, reinforcement-learning
icyChessZero 中国象棋alpha zero
这个项目受到alpha go zero的启发，旨在训练一个中等人类水平或高于中等人类水平的深度神经网络，来完成下中国象棋的任务。目前这个项目仍在积极开发中，并且仍然没有完成全部的开发，欢迎pull request 或者star。 然而受到计算资源限制，这样庞大的任务不可能在一台机器上完成训练，这也是我完成了分布式训练代码的原因，希望各位小伙伴能够加入，一起训练这样一个中国象棋alpha go的网络。
我的估计是达到4000～5000elo分数的时候深度网络可以达到目标，现在深度网络已经到了3000分的边缘，达到人类中上水平的目标并不是不可能的。
目前的elo：
详细胜率表：
当然，目前棋力还比较一般，因为是从完全随机开始训练的，比方说某个对局片段(800 playouts)：
推荐代码运行环境
- python==3.6.0
- tensorflow==1.4.0
- threadpool==1.3.2
- xmltodict==0.11.0
- urllib3==1.22
- numpy==1.14.3
- tflearn==0.3.2
- pandas==0.19.2
- scipy==1.1.0
- matplotlib==2.0.0
- tqdm==4.19.4
- tornado==4.5.1(集群master必须安装，slave不需要)
- uvloop==0.9.1(windows可不装)
推荐机器环境
- 机器在北京邮电大学机房中（不加入我们的集群或者自组集群可以不管这条）
- windows 或 linux服务器
- 16G以上内存
- 4个以上CPU物理核
- 一张以上能够进行深度学习运算的显卡
加入我们的集群训练(北邮校内only)
目前我们的集群已经有四台gpu机器(两台windows，两台linux)在夜以继日地运行，我们需要更多，如果你恰好有北邮机房内闲置的gpu服务器的权限，希望你加入我们，一起训练中国象棋的alpha go zero。
集群分为master和slave，加入集群的机器均为slave，master和slave分工如下：
- slave : 负责自动从master拉取最新模型权重，完成自对弈，并且把棋谱自动上传到master
- master: 负责给slave提供权重，并且负责模型的更新，评估，以及从slave接收棋谱
如果你想要加入我们的训练：
- 首先联系我 qq/微信： 892009517 ;邮箱：[email protected], 由于项目仍然在快速迭代，所以经常需要更新代码，与我联系获得最新的消息或代码更新的时间很必要。
- 如果你实在不想联系我，那么可以进行如下操作，直接加入集群(不推荐)：
windows 机器加入集群(北邮校内only)
clone工程后在cmd中执行下面命令
cd script
./multithread_start.bat [thread_number] [gpu_core] [python_env]
比如:
./multithread_start.bat 10 0 python3
意味着在0号GPU上用python3环境跑10个进程（一般一个1080ti GPU可以支持到至多24个进程），然而GPU并不是唯一瓶颈，不推荐跑超过物理核数两倍的进程数量。
linux 机器加入集群（北邮校内only）
clone工程后在shell中执行
cd script
比如:
sh multithread_start.sh -t 10 -g 0 -p python3
意味着在0号GPU上用python3环境跑10个进程（与上面windows版本对应)
总之
总之，如果想要加入集群，请先联系我，因为如果直接加入集群，可能会有我这边代码更新了然后部分slave没更新的情况，这样会造成不一致，后果会不可知。
自组集群
如果你在北邮没有机器，然而在校外有一些机器，希望能跑起来这样一个分布式程序，那么请按照下面的步骤做：
- 确定你要这么做，这是一个耗时，昂贵，不讨好,但是有点意思的工作
- 推荐你的机器（们）的环境满足推荐配置，并且安装好应该装的包
- master 机器一定要是linux（目前没有支持master也是windows）
- fork一份icyChessZero的代码,找到
config/conf.py这个文件，把server的ip改成你希望的master的ip - master 和slave分别clone这份fork的代码
- 在master上
cd scripts运行initize_weight.py生成第一份随机权重 - 在master上
cd distribute运行distributed_server.py开启master服务端口 - 在slave机器上起slave进程的方法同上文"加入集群"
- master上如果有空闲的资源可以起几个slave进程
- 模型更新和validate的方法在scripts/daily_update.sh中，按照你的需求改这个shell文件，并且把它放到crontab中设置为每小时运行一次（它会检查棋谱数量，数量足够后它会执行模型更新和评估工作）
查看棋谱
slave机器运行出来的棋谱在 data/distributed 目录下，是cbf文件，可以通过"象棋桥"软件查看，也可以在 ipynbs/see_gameplay.ipynb 中查看
查看训练状态
master 机器可以在ipynbs/elo_graph.ipynb 中查看集群训练的模型的elo到什么水平了。
没做的事
还有挺多东西可以做的，工程也还在快速开发,比如：
- 给棋谱加上一些meta，比如每一步的mcts分析，方便查个别case
2.长将和长捉的判断还没有做
- 给代码加上版本限制，master只接受与自己版本相同的slave的棋谱
- 专门搞一个web ui实时展示elo和棋谱等
- readme写清楚模块划分 .....
等等等等 如果你发现有你想做的，提提pull request或者联系我撒
这个work的一些细节已经以草稿的形式发布在了： http://icybee.cn/article/69.html
联系方式：
- QQ/wechat : 892009517
- 邮箱 : [email protected]
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].