Hurtig modstandermodellering i simplificeret poker

Jensen, Flemming

2. term

Computer Science, Master

2009

2009-06-10

134 pages

In poker, a game-theoretic optimal solution offers a strong strategy against unknown opponents. However, against opponents that do not play optimal in the game-theoretic sense, using opponent modeling to detect and exploit weaknesses in the opponent's play can yield better results. In this thesis we explore the feasibility of two distinct opponent modeling techniques in the context of a simplified version of heads-up Fixed Limit Texas Hold'em called Leduc Hold'em. Since many real life poker games often last for only a few hundred rounds, we focus on the feasibility of opponent modeling in games that last for at most 500 hands. In the first technique, which we call explicit opponent modeling, we compute a best-response strategy using an estimation of the opponent's strategy obtained through available observations. In the second technique, which we call implicit opponent modeling, we focus on, amongst a fixed set of candidates, selecting a strategy that is best against the opponent. By implementing a test framework, we are able to conduct a wide range of experiments. Based on the experimental results we conclude that, in our setting, the explicit modeling approach appears infeasible, while the implicit modeling approach appears feasible. We end this thesis by outlining a new opponent modeling technique that will be implemented and tested in the near future.

Poker ; LIMIDs ; Game theory ; Opponent modeling ; Spilteori ; Modstandermodellering

Download
View record in AAU Student Projects