Design And Implementation Of Self-play Chess Game Learning Example Generator

Computer games is an important research direction in the field of artificial intelligence.As an important method in the study of computer chess games,self-play learning refers to a learning method that depends only on the process of the game and the outcome of winning or losing.Throughout the self-play process,no domain knowledge was set in advance,except for the rules of playing chess,and no experts participated in the guidance.At present,based on the maximum and minimum algorithm,the ?-? pruning algorithm and the Monte Carlo Search algorithm,the self-play game learning has achieved excellent results,but there are still many clear problems in the research of artificial intelligence in chess games.Among them,one of the main problems researchers need to face is how to collect as many high-quality learning examples as possible within a limited time.In response to this problem,this paper designs and implements a self-play game learning example generator software system.After fully analyzing the researchers' needs for the sample generator,two subsystems are designed and completed: the tic-tac-toe learning sample generator subsystem and the checkers-learning sample generator subsystem.The main work of this article is as follows:?The conceptual framework of the sample generator is analyzed.The framework includes four parts: generator,execution system,appraiser and generalizer.The execution system takes the new chess game generated by the generator as an input,self-play according to the evaluation function V and obtains learning examples(moving records),and the evaluator uses the learning examples to train.?Designed and implemented the tic-tac-toe and checkers learning sample generator subsystem.The tic-tac-toe chess subsystem exhausted the chess surface to generate all learning examples,and visually showed the complete game tree andsample generation process;the checkers subsystem is complex The degree is higher,and the learning example is obtained by using the method of side-to-side game preservation.?Two chessboard state scoring methods and two learning indicators are proposed to test the generating ability of the two subsystems respectively.The results show that both subsystems have reached the design goals.The tic-tac-toe subsystem can generate effective samples under both evaluation methods,and the checkers subsystem can control the number of samples generated according to the indicators.The self-play learning sample generator software system has been provided to related researchers as a platform.The operation in the past half year shows that the system is running normally.It can effectively generate learning examples of two types of chess games,and satisfies the needs of researchers' experimental verification well.