Q-Learning Tic-Tac-Toe Bot

This project implements Q-learning reinforcement learning to play Tic-Tac-Toe against itself to learn optimal strategies.

Features

• Q-Learning table mapping states & actions with rewards
• Self-play training of 100k+ rounds
• Evaluation & training functions
• 50% reduced training time due to canonicalization of game states

How It Works

• Board is represented as a tuple
• Each unique state is an entry in the Q-table
• Uses epsilon-greedy exploration, with epsilon decaying throughout training

Installation & Training

# Clone the repository
git clone https://github.com/DanielJoy6/TicTacToe-Bot.git
cd TicTacToe-Bot

# Install dependencies
pip install numpy

# Train the agent
python train.py

Future Improvements:

• Reducing number of boards in Q-table by mapping rotations into single state
• Saving & Loading q-table
• MatPlotLib plots of winrate over time throughout training

For any questions, please email me at danieljoy2345@gmail.com