对称无关的模式编码方法在定式学习中的应用

A pattern encoding method independent of symmetry applied in joseki learning

计算机围棋中的定式可以从棋谱中学习,采用与模式对称性无关的编码方法可以取得更好的储存效果.这是一种基于模式的邻近特征、轮廓特征和行列特征的模式编码方法,与模式的平移、旋转、翻转及黑白对称性无关.实验结果表明其哈希性能良好,模式间的查找迅速,也可以在官子和死活问题中得以应用.

Joseki can be learned from game records in Computer Go. To more effectively maintain the joseki library, a pattern encoding method independent of symmetry is needed. This is provided by an encoding method is designed based on the pattern＇s neighbor character, the outline character and the row and column＇s character. The results show that this method＇s hashing performance is good, and it rapidly switches between patterns. The method can also be applied to tsumego and yose in Computer Go.