Extracting Important Patterns for Building State-Action Evaluation Function in Othello

Abstract

Important patterns are necessary in building a state-action evaluation function in board game. A pattern is good if its features are informative. In this paper, we propose a method to evaluate whether a pattern is informative. From that, important patterns can be extracted automatically. Patterns must be extracted from high quality and huge set of game records. In experiment, we implement the method in such set of Othello game records.