In the current assignment we examine two methods for the computation of equilibria in zero sum games. We demonstrate their theoretical foundations and we test them on several games. We illustrate their behaviour regarding their convergence and we compare them to each other.
The first one is the Fictitious Play, a prominent model-based learning rule (implementation in fictitious_play.py) and the second one is Reinforcement Learning, in a model-free approach (implementation in minmax_Q_RL.py).
The structure and content of this assignment is mostly inspired from the book Multiagent Systems: Algorithmic, Game-Theoretic, and Logical Foundations of Yoav Shoham and Kevin Leyton-Brown.
The report is found in Report.pdf.