Matchbox Computer

Overview

This project implements a simple machine learning approach using the Matchbox (MENACE - Machine Educable Noughts and Crosses Engine) concept to train a computer agent to play Tic-Tac-Toe. MENACE is an early form of machine learning applied to the game of Tic-Tac-Toe, where matchboxes and beads are used to simulate learning by reinforcing winning moves and punishing losing ones.. The program pits a random agent (playing as "X") against a computer agent (playing as "O"). The computer agent learns by reinforcing winning moves and punishing losing ones.

Features

• Train a computer to play Tic-Tac-Toe against a random agent.
• Computer improves its performance over time by learning from wins, losses, and draws.
• Visualize performance trends using matplotlib plots.

Prerequisites

Before running the code, ensure you have the following installed:

• Python 3.x
• Required libraries:
  • itertools
  • random
  • pickle
  • matplotlib

To install the necessary dependencies, run:

pip install matplotlib

Running the Program

1. Clone the repository:

   git clone <repository-url>
   cd <repository-directory>

2. Run the training: To start the training process, execute the following command in your terminal:

   python <script_name>.py

   The script will simulate a series of Tic-Tac-Toe games between the random agent and the computer. After every num_games rounds, the script will display statistics about wins, draws, and losses. You will have the option to continue training after each batch.

3. Save and Load Progress: The state of the game and the computer's learned strategy is saved in a data.pkl file. This ensures that the computer continues to improve even after the program is restarted.

4. Visualize Performance: After the training session, a plot showing the total wins, losses, and draws over time will be displayed.

Explanation of Key Concepts

• Matchboxes: Each unique Tic-Tac-Toe board state is stored in a matchbox. The beads in each matchbox represent the available moves for the computer at that state.
• Reinforcement Learning: The computer is rewarded by adding beads for winning moves and penalized by removing beads for losing moves. This way, the computer learns which moves lead to better outcomes.
• Training: During training, the computer repeatedly plays games against a random opponent. After every batch of games, statistics are recorded, and the user is prompted to continue training or stop.

Training Parameters

• TRAINING_EPOCHS: The number of games to play in one batch of training. Default is set to 10000.
• INITIAL_NO_OF_BEADS: The number of initial beads for each move in an empty board.

Customization

You can modify the TRAINING_EPOCHS and INITIAL_NO_OF_BEADS to adjust the length of training and initial randomness of the computer's moves.

License

This project is open-source and available for use under the MIT License.