Using Partial Order Bounding in Shogi

Grimbergen, R., Hadano, K. and Suetsugu, M. (2003)

in: The 8th Game Programming Workshop in Japan (GPW2002), pp. 52--58, Kanagawa, Japan.

Abstract

Using a weighted sum of feature values as evaluation function has some important drawbacks. If an evaluation function with a partial order of features is used, information which is otherwise lost by using a weighted sum can be preserved and better move decisions can be made. Partial Order Bounding (POB) is a new search method that promises to be more efficient than earlier attempts to search with partial order evaluations. Thus far, POB has only be applied to Go endgames. In this paper, we discuss the use of POB in shogi. We outline the implementation decisions that have to made when using POB. These concern the choice of the partial order evaluation, how to set the search targets in POB and how to manage multiple iterations of POB. We compare the results of a number of different implementation schemes of POB in shogi. The best scheme is able to find the correct move in 27 out of 50 next move shogi problems, using an average time of 48 seconds per problem. We consider this a good starting point for further research in Partial Order Bounding.