應用於象棋開局庫之工作層級AB-DUAL*搜尋演算法

摘要

Constructing a passive opening book for Chinese Chess requires the collection of thousands of expert games played on the Internet and filtering and ranking of all positions in the opening book based on factors such as the number of wins/draws/losses. A major issue here is the consistency between the opening book and the game-playing program. That is, the “statistical good” positions could be “weak spots” for game-playing program. In this paper, we evaluate all positions with game-playing program under job-level system to speed up the computation and maintain the consistency for the construction of the Chinese Chess opening book.Generic job-level search was proposed to solve computer game applications by dispatching jobs to remote workers for parallel processing. This approach leverages game-playing programs and encapsulates them as jobs. Such an approach is well suited for a distributed computing environment, since these jobs can be run independently by remote processors in a job-level system. This paper applies job-level search to AB-DUAL*, which is an extension of Alpha-Beta Search but uses zero-window to increase pruning. In our experiments, the results demonstrate significant performance improvement and speedups.