ChessWarrior

Final Project of Spring 2018 Artificial Intelligence at Tongji Univ.

A Chess AI based on Alphago and Alphago Zero

Currently finished supervised learning with alpha-beta search.

A pytorch version can be found here alpha-chess

Demo

ChessWarrior(white) vs. Stockfish(balck; elo 1600)

Usage

There are three modes: data train and play

• data: generate processed training data
• train: the main process to train neural network
• play: use the best model to play chess

python run.py -mode data

python run.py -mode train -lr 0.01

python run.py -mode play --no-cuda

Requirements

environment windows/linux

python 3.x

• tensorflow-gpu
• keras==2.1.5
• h5py
• numpy
• python-chess

just

pip install -r requirements.txt

Todo

• Reinforcement Learning
• MCTS

Pretrained models

https://pan.baidu.com/s/1C2FVG9vQds-odrfgw5bUGQ passwd: imhk