Computer Go Game Research Based On Monte Carlo Tree Search

In recent years, with the rapid development of computer software, hardware and Internet, meanwhile, the related disciplines are increasingly boosted. Artificial Intelligence has long been a crucial topic, and computer game which draws wide attention is one of its research directions. Considered as one of the most complex games, the challenging go can adequately testify the level of development in the artificial intelligence. Therefore, the study concerning computer go game plays a significant role in computer game, artificial intelligence and even the field of computer science.In response to the difficulties of summarizing the expertise of go and excessive searching space, traditional theories and methods of computer game has been both rarely applied to practice so far. In the past few years, computer go game employs Monte Carlo method to make dynamic evaluation and introduces UCT(Upper Confidence Bound Applied to Trees) on the basis of great advancements in the performances of computer, which markedly enhance the efficiency of searching and the performances of evaluation. Furthermore, the level of the computer go game has attained great improvement. In addition, the clustering algorithm and intelligence are not only applied to the game of go, but also applied to other problems such as planning and decision-making problems. Thus, the research findings in computer go game is undoubtedly of considerable realistic significance.Combined Monte Carlo method with UCT, MCTS (Monte Carlo Tree Search) is a popular and better method in advanced computer go program at present. However, this method still has some drawbacks. Based on the thought and features of MCTS and some disadvantages of Monte Carlo method and UCT, the thesis proposes two improvements of absolute pruning strategy and progressive widening strategy and demonstrates the correctness and necessity of these improvements in theory. Besides, this thesis chooses Fuego, an advanced open source go program which adopts MCTS, as the improvement test program, and makes the improved Fuego and the original one to play respectively with another advanced program called Pachi through a platform named GoGui.Then, this thesis testifies and analyses the time and results of all the test games. The research findings indicates that Fuego’s winning percentage is slightly increasing and the time reduces, which demonstrates the improvements has a practical effect and it is of certain practical value.