Hand Strength Forecast Model And NPC Design For Texas Hold’em Poker

Artificial Intelligence in game is a hot research field. This paper an analyzes of the present research of artificial intelligence technology in poker games, and pays more attention to the analysis of artificial intelligence technology in incomplete information card games, specifically in Texas Hold’ em Poker.The main contributions of our work can be summarized as following:Firstly, based on classification methods in data mining, we design a model strategy of hand strength forecast. With a specially designed feature vector using combined data from both the current hand and the history hands, our designed model is more simple and effective compared with the existing models.Secondly, based on Naive Bayes, Support Vector Machine and Bayesian Networks algorithms, we implement three classifiers for hand strength forecast. Using these classifiers, we develop three corresponding NPCs. We carry out a round robin format comparison test of our three NPCs with three other public NPCs. The comparison test results show that these three NPCs based on our hand strength forecast model is effective, and the NPC using the Bayesian Networks algorithm has a good performance both in basic combat ability and adaptability.Thirdly, based on the protocol designed by ACPC(Annual Computer Poker Com-petition), we develop a fighting test system for artificial intelligence research on Texas Hold’em Poker. This system can be used not only in human-computer fighting but also human-computer fighting tests. It can test the proposed hand strength forecast algorithms and make real-time control of the game to watch the game status.