Quixo Game using Reinforcement Learning

How to read this repo?

First I invite you have a look to draft.pdf where it is a scientific research draft article where it contains all information.

Main files:

• env_previous.py: Env that uses previous trained version of it
• env_random.py: Env that uses a random player
• game.py: The game functions and play methods
• train_*.py: Training Scripts
• tune_*.py: Tuning scripts
• utils.py: Necessary functions to connect between the game and the RL env (encode and decode actions)
• test.py: Testing the trained agent
• env_checker.py: A script to check the configuration of the env.

Secondary files:

• Quixo.pdf: An explanation of the game
• requirements.txt: The python env libraries and packages
• quixo_ppo_random_opponent_*.zip: PPO trained models.
  • 2M, 2M_previous, longest are the first, second and third trainings respectively.

Non-used files:

• easyAI.py: An non-complete implementation of Negamax for Quixo.
• mcts.py: An non-complete implementation of Monte-Carlo Tree Search for Quixo.

---

This work has been done as a project for the Computational Intelligence course 23/24 @ PoliTO.

https://github.com/squillero/computational-intelligence