This repository implements code for magnetic mirror descent, as described in the paper A Unified Approach to Reinforcement Learning, Quantal Response Equilibria, and Two-Player Zero-Sum Games.

To install the package, run the following code:

git clone https://github.com/ssokota/mmd.git
cd mmd
pip install .

Four example scripts for magnetic mirror descent are included in the examples directory:

1. examples/nfg/perturbed_rps.py is a script for computing QREs for perturbed RPS
2. examples/efg/aqre.py is a script for computing AQREs in extensive-form games
3. examples/efg/nash.py is a script for computing Nash equilibria in extensive-form games
4. examples/h2h/evaluate.py is a script for head-to-head evaluation in extensive-form games

There is also an additional example script examples/efg/cfr.py for running CFR in extensive-form games.

The reference for the paper is:

@inproceedings{
sokota2023a,
title={A Unified Approach to Reinforcement Learning, Quantal Response Equilibria, and Two-Player Zero-Sum Games},
author={Samuel Sokota and Ryan D'Orazio and J Zico Kolter and Nicolas Loizou and Marc Lanctot and Ioannis Mitliagkas and Noam Brown and Christian Kroer},
booktitle={The Eleventh International Conference on Learning Representations },
year={2023},
url={https://openreview.net/forum?id=DpE5UYUQzZH}
}