Author: Paul Clarke
Date: 10:04:43 06/28/05
Go up one level in this thread
On June 28, 2005 at 06:39:35, Tord Romstad wrote: >On June 28, 2005 at 03:26:12, Reinhard Scharnagl wrote: > >>10x8 chess also compatibly is leaving the old ways of chess programming >>moreover having about 25% more moves in each ply of computing, which might be >>a good way to approach to Go programming later. > >I used to think so, too, but my experience with hexagonal chess has made me >change my mind. My program plays both games, using the same source code >(with a few tiny differences). The games are similar, except that the average >number of legal moves is about 2.5 times bigger in hexagonal chess. I expected >this to be a major problem, but it turns out that it isn't. The search >techniques >from classical chess work just as well in the more complex game of hexagonal >chess. Of course the bigger branching factor makes it impossible for the >program to search quite as deeply as in classical chess, but this is a problem >for human players as well. > >The really hard thing about go and shogi compared to chess (from a >programmer's point of view) is the difficulty of writing a good evaluation >function. In chess, material is much more important than everything >else. You can play chess well without doing much more than counting >material. In go or shogi, you will have to work really hard to produce >an evaluation function which works as well as a material-only eval in >chess. Argueing for branching factor being a problem, the strong shogi programs for which I've seen descriptions all use some form of pruning beyond alpha-beta with null move: Gekisashi uses realisation probability pruning, while YSS and Spear use plausible move generators. I can think of a few possibilities as to why this seems to differ from your experience with hexagonal chess: 1) It might just be selection bias: other strong shogi programs might be using the standard techniques used in chess but not writing about them precisely because the techniques are standard. 2) The branching factor problem is potentially worse in shogi: positions with 200+ moves aren't out of the ordinary. 3) The standard of human play in shogi is, I imagine, higher than that in hexagonal chess. If you had to compare your program's play to a hexagonal chess equivalent of Habu, you might decide you needed more pruning :-).
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.