Reinforcement learning applied to Texas Hold'em Poker

Introduction

We propose a Q-learning approach to solve a poker-based game, i.e single-round all-in/fold poker. We also implemented a multi-round approach also based on Q-learning as well as a Deep Q learning for exploratory purposes. This work as been done as part of our final project for the Reinforcement Learning course of CentraleSupélec.

Environment

We want to kindly thank Ferdinand Schlatt for developing open-source poker environment package clubs_gym on which our implementation is based, and for his quick answers to all issues we may have had with his package.

Structure

│   .gitignore
│   README.md
│   requirements.txt
├───notebooks
│   ├───agents.py
│   ├───all_in_fold_several_hands.ipynb
│   ├───all_in_fold.ipynb
│   ├───base_nb.ipynb
│   ├───all_in_fold_deep_q.ipynb

Implementations

The single round approach is available in the all_in_fold.ipynb notebook and the multi-round approach is available in the all_in_fold_several_hands.ipynb notebook, and the Deep_Q is in all_in_fold_deep_q.ipynb.