Computing card probabilities in Texas Hold'em

Abstract

Developing Poker agents that can compete at the level of a human expert can be a challenging endeavor, since agents' strategies must be capable of dealing with hidden information, deception and risk management. A way of addressing this issue is to model opponents' behavior in order to estimate their game plan and make decisions based on such estimations. In this paper, several hand evaluation and classification techniques are compared and conclusions on their respective applicability and scope are drawn. Also, we suggest improvements on current techniques through Monte Carlo sampling. The current methods to deal with risk management were found to be pertinent concerning the agent's decision-making process; nevertheless future integration of these methods with opponent modeling techniques can greatly improve overall Poker agents' performance.