Using master games statistics on the squares played

IPSJ

Description

Most state-of the art Shogi programs adopt variants of the fixed-depth alpha-beta search algorithm. However, there are numerous search extensions that explore "interesting" lines of play more deeply. Most search extensions make the decision to extend a certain line of play dynamically. We propose a static topological frequency based search extension. That is, the decision to extend searching of a certain move is dependant on the frequency the relevant square was played in master games. As this method was inspired by and akin to Realization Probability Search we named it Topological Frequency Bias (TFB). The proposed TFB has been incorporated to our computer Shogi program, TACOS, and two experiments proved its effectiveness.

Most state-of the art Shogi programs adopt variants of the fixed-depth alpha-beta search algorithm. However, there are numerous search extensions that explore "interesting" lines of play more deeply. Most search extensions make the decision to extend a certain line of play dynamically. We propose a static topological frequency based search extension. That is, the decision to extend searching of a certain move is dependant on the frequency the relevant square was played in master games. As this method was inspired by and akin to Realization Probability Search we named it Topological Frequency Bias (TFB). The proposed TFB has been incorporated to our computer Shogi program, TACOS, and two experiments proved its effectiveness.

Journal

Details 詳細情報について

Report a problem

Back to top