Planning to Guide Opening and Middle Game Play in Shogi

Grimbergen, R. and Matsubara, H. (2000)

Proceedings of the 42nd Japanese Society for Artificial Intelligence Workshop, pp. 19--24

Abstract

In shogi the use of an opening database is not as effective as for example in chess. Tests with a large opening database show that on average only move suggestions for the first nine moves can be expected. Therefore, planning the development of pieces in the opening must be guided in a different way to avoid that the program will be at a disadvantage early in the game. In this paper we propose two methods for partially matching the game position with positions in the database. The first partial matching method is half matching, in which only the pieces of the player to move are matched to the positions in the database. The second type of matching is quarter matching, where the pieces of the player to move in two 5x5 sections of the board are matched to positions in the database. We show that with these matching methods, we can extend the phase of the game in which the opening database is effective. We will also show that these partial matching methods improve the overall level of play of a shogi program. The test results also suggest that quarter matching combined with half matching is better than only using half matching.