Pseudo-Optimal Strategies in No-Limit Poker

Games have always been a strong driving force in artificial intelligence. In the last ten years huge improvements have been made in perfect information games like chess and othello and the strongest computer agents can beat the strongest human players. This is not the case for imperfect information games such as poker and bridge where creating an expert computer player has shown to be much harder. Previous research in poker has either adressed fixed-limit poker or simplified variations of poker games. This paper tries to extend known techniqes successfully used in fixed-limit poker to no-limit. Nolimit poker increases the size of the game tree dramatically. To reduce the complexity an abstracted model of the game is created. The abstracted model is transformed to a matrix representation. Finding an optimal strategy for the abstracted model is now a minimization problem using linear programming techniques. The result is a set of pseudo-optimal strategies for no-limit Texas Hold’em that perform well as long as the players’ stack sizes are fairly small.