/Selfplay_Gomoku

impala selfplay gomoku

Primary LanguagePython

INTRODUCTION

This repo implement an Gomoku AI by selfplay, the main code is modified from seed-rl.

I have trained an model that got 90% winrate vs rule AI (from an open source repo).

winrate to rule AI

Training info: RTX2070 / I7-9700 / 2~3day / 2G-frame-sample

Download Checkpoint

HOW TO USE

1. install nvidia-docker

https://github.com/NVIDIA/nvidia-docker

2. build docker container

cd docker
bash build.sh

3. train

bash train_local.sh

Accessing Tensorboard Monitor during the training: localhost:6006.

4. try to play with it

img

bash play_local.sh #CHECKPOINT PATH(eg: 1588068127/ckpt-124)#