一种围棋定式的机器学习方法

A Machine Learning Method of Joseki Database for Computer Go

提出了一种围棋定式的机器学习方法。利用此方法可实现从棋谱库中自动提取定式并生成定式库。此外，对于棋谱数量较大的情况，采用分阶段学习方法，提高了学习效率。应用此方法，对34 000局棋谱进行处理，得到定式点680 638个。最后，还给出了1种基于组合博弈理论在计算机围棋博弈系统中使用定式的方法。

A machine learning method of Go joseki for computer Go game system is proposed in this paper. By using this method, various Go joseki can be automatically extracted from human-played Go game files so as to construct a joseki-tree database. The learning procedure can be separated by several steps when the learning data set is too large. This approach improves the learning efficiency. 34 000 human-played Go games are processed and 680 638 joseki nodes are created. In addition, a method of how to apply joseki, which is based on combinatory game theory, to computer Go game system is presented.