Source code from paper (AIStat 2017): https://arxiv.org/abs/1606.08718
bibtex: @inproceedings{perolat2016learning, title={Learning Nash Equilibrium for General-Sum Markov Games from Batch Data}, author={P{'e}rolat, Julien and Strub, Florian and Piot, Bilal and Pietquin, Olivier}, booktitle={Artificial Intelligence and Statistics}, year={2017} }
Requirement:
- tensorflow > 1.0
- numpy
- pickle
Given a random generated markov games with N players, this code computes the players'stategie that would tend to an epsilon nash-equilibrium. Parameters are defined in the main.py for a two plyer general sum-games.
We are aware that the code is not very clean (bad design paterns, hardcoded conf, few comments), I would be happy to answer your questions and update the code accordingly. Feel free to contact me for more details!