A method of computing winning probability for Texas Hold'em poker

Abstract

Texas Hold'em poker is one of the most popular card games in computer games. Most of the times hand evaluation is necessary for computing hand strength which is also called winning probability. We need to traverse all combinations of cards to get the accurate hand strength in conventional approach. Usually Monte Carlo simulation or look-up table could reduce the computation. To improve efficiency, this paper propose an algorithm for poker hand classification and introduce linear regression to calculate the approximate effective hand strength. Firstly getting the classifying result of 5 cards or 6 cards, then generating feature vector through data preprocessing, at last the predictive value of the effective hand strength calculated by the linear regression equation. The experiment results show that the average absolute error between the predicted value and the label is 0.034. Compared with the traditional evaluation approach, our method is more effective to computing winning probability without lookup table.