Portal de Periódicos da CAPES

Auto-encoder neural network based prediction of Texas poker opponent's behavior

2021; Elsevier BV; Volume: 40; Linguagem: Inglês

10.1016/j.entcom.2021.100446

1875-953X

Shuai Wang, Hao Wang, Qiang Gao, Lina Hao,

Video Analysis and Summarization

Imperfect information game is the difficulty in the game field. In the study of imperfect information games, an accurate and effective opponent modeling method is essential. Texas Hold 'em is a typical game with imperfect information. The main form of opponent modeling is to predict the opponent's behavior. This article takes Texas Hold 'em as the specific research object and proposes an opponent's behavior prediction method based on an auto-encoding neural network. We use the International Computer Poker Tournament historical data as experimental data to model Texas Hold 'em poker players' behavior and train an auto-encoding neural network model that can be applied to different stages of the game. In the article, the auto-encoding network model's structure and parameters are experimentally studied, and the quantized conjugate gradient convergence algorithm (SCG) and softmax classifier are used. Experimental test results show that this method can accurately predict the opponent's behavior information in a short time, which will provide precious help for the decision-making of the Texas Hold 'em agent.