Study On The Algorithm Of Limited Bet Texas Leduc Hold ’em Based On Reinforcement Learning

As an important branch of artificial intelligence,Machine game has attracted more and more scholars’ attention.As a popular competitive poker game,Texas Hold’em has always been a hot issue in the field of machine game research.With the advent of Alpha Go,the problem of machine games with complete information has been basically solved,but incomplete information machine games such as Texas Hold’em have not yet been solved well due to the complexity of state space.Therefore,it is of great theoretical and practical significance to study the incomplete information machine games such as Texas Hold ’em.This thesis uses reinforcement learning algorithm to study Texas Hold’em.First analyze the Counterfactual Regret Minimization algorithm and the algorithm flow,using Nine Barrels Technology Combined with Public Cards extraction technology to improve the original card abstract way,each node of the hand and public brand portfolio allocation,adjust the brand strategy according to the dynamic changes of the public brand,in order to reduce the state space complexity and improve game performance;On this basis,the game state and agent training mode of CFR algorithm are analyzed,and the implementation strategies of CFR algorithm in two poker games are given.Second Q learning algorithm and the algorithm of reinforcement learning process is analyzed,and combining with the valuation function using dynamic random sampling to estimate of Texas poker force,build the Q learning algorithm in Texas poker model,through the analysis of Q learning algorithm of the game state and agent training methods,Q learning algorithm is given in the implementation strategy of the two kinds of poker game.Will finally proposed algorithm applied to Texas poker and whose poker,design program and trained two game of CFR agent and Q learning agent,agent respectively after training and rule-based expert play against the AI,and get the CFR algorithm according to the results of the experiment and the Q learning algorithm in different complexity of the algorithm results in complete information game and characteristics.The improved algorithm can provide a reliable and stable strategy selection scheme for different poker games,and the improved NBTCC bottom card extraction technology can also show good performance in pure strategy testing.In addition,it can be seen from the Leduc poker comparison experiment that Q learning algorithm has a good effect on rule-based AI in simple games,which lays a foundation for the popularization and application of Q learning algorithm in machine games.