A Generalized Extensive-Form Fictitious Play Algorithm

Tim P. Schulze
2023-10-15
Abstract:We introduce a simple extensive-form algorithm for finding equilibria of two-player, zero-sum games. The algorithm is realization equivalent to a generalized form of Fictitious Play. We compare its performance to that of a similar extensive-form fictitious play algorithm and a counter-factual regret minimization algorithm. All three algorithms share the same advantages over normal-form fictitious play in terms of reducing storage requirements and computational complexity. The new algorithm is intuitive and straightforward to implement, making it an appealing option for those looking for a quick and easy game solving tool.
Computer Science and Game Theory,Artificial Intelligence
What problem does this paper attempt to address?
### What problem does this paper attempt to solve? This paper primarily explores how to find equilibrium solutions in two-player, zero-sum games. Specifically: 1. **Algorithm Introduction**: - Proposes a simple and intuitive Generalized Extensive-Form Fictitious Play (GXFP) algorithm for finding equilibrium in two-player, zero-sum games. - This algorithm is more efficient in implementation compared to the traditional Normal-Form Fictitious Play (FP) algorithm, reducing storage requirements and computational complexity. 2. **Algorithm Comparison**: - Compares GXFP with two existing algorithms: Extensive-Form Fictitious Play (XFP) and Counter-Factual Regret Minimization (CFR). - These three algorithms share the same advantages in terms of storage and computational efficiency, but GXFP is simpler and more intuitive to implement. 3. **Theoretical Proof**: - Proves that the GXFP algorithm is equivalent to a generalized fictitious play process and inherits the convergence properties. - Utilizes the concept of counter-factual utility to simplify calculations. 4. **Benchmark Testing**: - Uses the classic poker game model proposed by von Neumann and Morgenstern as a benchmark to demonstrate the performance of different algorithms. - Compares the results of undisturbed games (without minimum probability constraints) and disturbed games (with minimum probability constraints), revealing differences between various strategies. Through this research, the paper aims to provide an efficient and easy-to-implement algorithm suitable for solving large-scale game problems.