Tackling the UNO Card Game with Reinforcement Learning

Author: Bernhard Pfann
Article: https://towardsdatascience.com/tackling-uno-card-game-with-reinforcement-learning
Status: Done

Description

In this project I tried to analytically derive an optimal strategy, for the classic UNO card game. To do so, I structured my work as follows:

1. Creating a game engine of the UNO card game in Python from scratch
2. Obtaining game statistics from simulating a series of 100,000 games
3. Implementing basic Reinforcement Learning techniques (Q-Learning & Monte Carlo) in order to discover an optimal game strategy

UNO card engine: In order to train a Reinforcement Learning (RL) agent how to play intelligently, a fully-fledged game environment needs to be in place, capturing all the mechanics and rules of the game. Class objects for Card, Deck, Player, Turn and Game are defined.

Statistics from simulations: By running multiple simulations of the game, the following questions are being tackled:

• How many turns do games last?
• How big is the advantage of the player making the first turn?
• What are the most likely situations in the course of a game?

Application of Reinforcement Learning: In agent.py, I defined the algorithm for a Q-Learning or Monte-Carlo agent, both working with a discrete state-action matrix. In config.py the preferred algorithm can be selected together with its main tuning parameters. Finally the results after training the RL-model are being analyzed in terms to cumulative win-rate and obtained Q-values.

Repository Structure

• assets/ collection of .csv files that are generated by execution of run.py
• notebooks/ analysis of simulated games
• src/ core package to simulate games
• config.py configurable parameters
• run.py executation file

Execution Instruction

Clone repository via HTTPS:

$ git clone https://github.com/bernhard-pfann/uno-card-game-rl.git

Install requirements and execute main file:

$ pip install -r requirements.txt
$ python run.py