Can Online Opponent Exploitation Earn More Than the Nash Equilibrium Computing Method in Incomplete Information Games with Large State Space?
Zhenzhen Hu, Shaofei Chen, Peng Li, Jiaxing Chen, Jing Chen · 2023
In incomplete information games with large state space like Texas Hold'em, Nash equilibrium computing and opponent exploiting are two ways for decision-making. While the Nash equilibrium computing way has achieved superhuman feats, the online opponent exploitation method has challenges in accurately modeling opponents online. In order to compare the Nash equilibrium computing method and the opponent exploitation method fairly for large state space incomplete information games, we take Texas Hold'em as an example, and propose a new online opponent exploitation method based on explicit opponent modeling, hidden cards inference, and winrate evaluation. Experiment results show that in matches against different opponents, a simple rule AI using the proposed method earns much higher incomes than the Nash equilibrium computing AI, indicating that the opponent exploitation method is a more effective way to play against opponents with weaknesses in large state space incomplete information games.