# Project Report: Tic Tac Toe with AI Opponents in Pygame 
## Research Questions This project aimed to answer the following questions: - Can I implement a simple Tic Tac Toe game using **Pygame**? - Can I make the game **visually appealing**? - Is it possible to integrate **different types of AI opponents**? ## Motivation The goal of this project was to improve my skills in **game development**, **visualization**, and **artificial intelligence** (AI). ## Implementation I developed a functional Tic Tac Toe game using **Pygame** and implemented **three different AI opponents**: 1. **Random AI** – chooses moves randomly. 2. **Q-Learning AI** – learns through reinforcement learning by playing against itself. 3. **Minimax AI** – calculates the best possible move by evaluating all future game states. ## Model Training For the reinforcement learning agent, I used **Q-learning**. The agent played **1 million games against itself**, during which it learned from over **800 unique game states**. ## Model Validation To validate the AI opponents, I: - **Played against each AI** myself, - Let the **AIs compete against each other**, - Published the results in a **Streamlit app**: [Open Streamlit App](https://georgvettergit-tic-tac-toe-reporting-apphome-zxc7r3.streamlit.app) The app also allows users to **play against the AI opponents**: - Play against **Random AI**: [Try it here](/assets/tic_tac_toe/random/index.html) - Play against **Q-Learning AI**: [Try it here](/assets/tic_tac_toe/qlearning/index.html) - Play against **Minimax AI**: [Try it here](/assets/tic_tac_toe/minimax/index.html) ## Results Evaluation The results showed a clear **increase in difficulty** among the AIs: - The **Random AI** is the easiest to beat. - The **Q-Learning AI** performs slightly better but still makes mistakes. - The **Minimax AI** is the strongest, as it always chooses the optimal move due to the limited number of possible states in Tic Tac Toe. ## Visualization - The game is **graphically displayed using Pygame**. - **AI vs. AI tournament results** are visualized in the **Streamlit app** using **Matplotlib charts**. ## Lessons Learned This was my **first hands-on experience** with reinforcement learning and designing **AI opponents**. Organizing a **tournament** between different AIs was also a new concept for me. Additionally, I deployed my **first Streamlit app**, gaining valuable experience in web-based visualization and app deployment.