Gomoku

Project Introduction

In this project, I implemented various AI algorithms for Gomoku, an extension of Tic Tac Toe. I have also completed a previous project on Tic Tac Toe, which can be found here: TicTacToe.

Set up Environment

To set up the environment for this project, follow these steps:

1. Install and setup conda.
2. Create environment.

conda create --name tictactoe python=3.8.19
conda activate tictactoe
pip install -r requirements.txt

Project Structure

• game.py: Contains the core game logic.
• gameplay.py: Manages game interactions between players, including both AI and human players.
• player.py: Defines an abstract class from which custom agents will be implemented.
• gomoku: Contains the AI agents that are to be developed and customized.
• bots: Includes pre-built Gomoku bots that can be used to evaluate the performance of agents.

Command-Line Usage

1. Example usage

python main.py -p1 master -p2 intermediate -n 4 -m ui

2. Options

python main.py -p1 [PLAYER_1] -p2 [PLAYER_2] -m [VISUALIZATION] -n [NUM_GAMES] -t [TIMEOUT] -s [SIZE]

• --player1 or -p1 : Choose player 1.

• --player1 or -p1 : Choose player 2.

  • Choices of player: 'minimax', 'alphabeta', 'mcts', 'qlearning', 'human', 'random'.

• --mode or -m : Choose visualization mode ('silent', 'plain', or 'ui').

  • 'silent' only shows game result (not possible for human player).
  • 'plain' shows the game state in terminal.
  • 'ui' shows the game with integrated UI.

• --num_games or -n : Number of games for evaluations. Not that players will be assigned 'X' and 'O' alternately between games.

• --timeout or -t : Set timeout for each AI move. No timeout is set for Human move. Default is 10 seconds per move.

• --no_timeout or -nt : No timeout for AI move.

• --size or -s : Size of the board for Gomoku (default=15).

• --load or -l : Load the Q table for Tabular Q-Learning agent, for weights for the features in Approximate Q-Learning agent. You should give only the file name of the weight to this argument.

• --no_train : No training for Q-Learning agents.

3. Evaluation

python evaluation.py -p [PLAYER] 

To evaluate AI agents using hand-crafted test cases, the output of the evaluation is saved in the evaluation folder.

• --load or -l: Loads the Q table for the Tabular Q-Learning agent or weights for the features in the Approximate Q-Learning agent. Only the file name of the weights should be provided for this argument.
• --no_train: Disables training for the Q-Learning agents.