Modeling Texas Hold'em Poker Strategies with Bayesian Networks
Konstantin Tretyakov, Liina Kamm · 2009
Poker is an exciting game of chance, psychology and skill. Development of an algorithm for playing poker (a pokerbot) has long been a beloved research topic for procrastinating computer scientists. In this work we join their ranks by presenting an example of a Bayesian network–driven approach to modeling strategy for a pokerbot. This approach is, naturally, not new – a well-known Bayesian poker player dates from 1999 [3] with improvements and variations being developed ever since [4, 5]. Nonetheless, we believe there are still numerous paths to be explored in this area. For example, it is only recently that the possibility of exhaustively enumerating all the 52!/43! = 1 335 062 881 152 000 nine-card combinations defining the states of two-player Texas holdem has become commonly accessible. Although the approaches we are going to discuss might be applicable, with small modifications, to a wide variety of poker games, we fix our goal here to the game of two player fixed limit Texas hold’em. Besides being one of the most popular poker variations in existence, it is also simple enough to fit nicely into the limited scope of this project.