Using pattern recognition and selective deepening to solve tsume shogi
Grimbergen, R. (1996)
In: Game Programming Workshop in Japan '96, pp. 150--159, Kanagawa, Japan
Abstract
In chess brute-force search methods have been very successful. However, games
like shogi have a game tree complexity that is much higher than chess.
Therefore, it
is unclear if the methods used for making strong chess programs will also be
successful for other games. In our research we will attempt to use
a more cognitively based method, pattern recognition, to build a strong shogi
playing program. In chess, pattern recognition has not been very successful,
but in this article we will take a different approach to patterns. We will
describe a more general definition of patterns and how we used this definition
and position evaluation to make a program to solve simple tsume shogi problems.
The results for solving five-ply and seven-ply tsume shogi problems are
encouraging. Furthermore, an evaluation function using only pattern recognition
turned out to be working better than an evaluation function using both
pattern recognition and position evaluation.