AlphaZero

AlphaZero and MCTS algorithm applied to game of TicTacToe and ConnectFour

(Image used for thumbnail purposes only)

Starting game of ConnectFour(rows=6, cols=7, in_a_row=4) with 10 rounds

============ Results ============

| Agent                | Wins | Draws |
|----------------------|------|-------|
| MCTSAgent_1          | 0    | 0     |
| AlphaZeroAgent_2     | 10   | 0     |

This is a concise and from scratch implementation of AlphaZero algorithm applied to game of TicTacToe and ConnectFour for the purpose of learning and experimenting with the algorithm.

Usage

Run the play_game.py file to play a game with settings defined in config/players.yaml file.

python play_game.py

Features:

Choose any 2 Players from following list by uncommenting any of the player_1 and player_2 settings in config/players.yaml:

• Random
• MCTS
• Human
• AlphaZero

Select a game by uncommenting game: TicTacToe or game: ConnectFour

Run tournament of N rounds by setting num_rounds: N

To view the game in action, set show_games: true

To view the end state of the game, set show_end_state: true

View a summary of the tournament after N round.

Training

Set the training parameters in the config/training_conf.yaml file. and start training using following command

python train_AlphaZero.py

Installation

1. Create new environment [optional]

   conda create -n alpha_zero python=3.10
   conda activate alpha_zero

2. Install supported version of PyTorch

3. Install requirements

   pip install -r requirements.txt

Acknowledgements

• DeepMind AlphaZero Papar - for the amazing paper and detailed explanation of the algorithm
• Simple Alpha-Zero blog - for their intuitive explanation and pseudo code
• AlphaZeroFromScratch - for their Easy to understand MCTS and AlphaZero implementations
• alpha-zero-general - for their implementation