realphongha / mnk-game-bot

Monte Carlo tree search for m,n,k-games (tic-tac-toe, gomoku...) bot with a Pygame UI.

mnk-game-bot

AI bot utilizing Monte Carlo tree search to play m,n,k-games (Tic-tac-toe, Gomoku...) with simple Pygame GUI for Player vs. Bot.

Implemented stuffs

Bot algorithms

• Monte Carlo tree search (MCTS) with Single-run parallelization:

@inproceedings{cazenave2007parallelization,
  title={On the parallelization of UCT},
  author={Cazenave, Tristan and Jouandeau, Nicolas},
  booktitle={Computer games workshop},
  year={2007}
}

Game configs

• Tic-tac-toe
• Gomoku 7x7
• Gomoku 9x9
• Gomoku 15x15 (MCTS seems not to be effective yet because of speed problem)

Cython optimization

• MCTS:

For end-game board checking
For getting possible moves on board
For MnkBoard and MnkState classes
For calculating UCB + score (not so big improvement)

How to install

First git clone https://github.com/realphongha/mnk-game-bot.git to clone this repo.
Go inside the repo: cd mnk-game-bot

Install requirements

Run:

pip install -r requirements.txt

Compile Cython file

Run:

python setup.py build_ext --inplace 

How to run

python play.py --cfg path/to/config_file.yaml

For example:

python play.py --cfg configs/tic_tac_toe.yaml