Quarto

An implementation of the Quarto game for Reinforcement Learning, with code to train and test several different agents (PPO, A2C, MaskablePPO).

Motivation

This project was completed in order to learn the basics of RL to play a simple game such as Quarto.

Built with

• numpy
• stable-baselines3
• sb3_contrib

Train instructions

python train.py --steps [N_STEPS] --algo [ALGO] [--path [PATH_TO_MODEL]]

To train using MaskablePPO, run:

python train_masked.py --steps [N_STEPS] --algo ppo_mask [--path [PATH_TO_MODEL]]

Test instructions

python test_model.py --algo [ALGO] --path [PATH_TO_MODEL] --episodes [N_EPISODES]

Quarto & MinMax with alpha beta pruning

Algorithm	Depth 1	Depth 2	Depth 3
	Time (s)	Nodes	Time (s)
-------------------	---------	-----------	---------
minMax	<0.01	256	0.02
Alpha-beta pruning	<0.01	256	0.02