Author: Fabien Letouzey
Date: 01:10:29 06/15/05
Go up one level in this thread
On June 14, 2005 at 16:03:42, Roger D Davis wrote: >Looks like the latest editions of Fruit and Fruit-Toga are very strong. Is there >any single structural feature that makes them so strong? Is it speed, or tuning, >or what? > >Roger Not a single feature, not speed and not tuning either. Like Volker said, the question is a hard one. I have not tried a different approaches yet, so unfortunately I cannot compare. In Fruit 1.0 I had a search mostly taken from my other game programs (Othello, international draughts, Awari, etc ...) and no eval (mostly PST). Afterwards I added the eval features that seemed to lack most in the games I watched. I also apply religious-like principles e.g. trying to keep parameter dimensionality low and keeping the search "robust" (tolerant to inconsistencies). However I picked these principles only for long term reasons and as yet they are not supposed to matter much. I believe that I could build strong engines with totally different approaches like attack tables or going for NPS. In my opinion, consistency is important but I cannot quantify or even measure it. I pick a design before writing the first line and try to stick to it. I learnt a lot by rewriting programs from scratch every 3-or-so years. I think it's hard to improve an old engine. Sorry that I can't really answer your question. I think only professional authors could, but won't. Fabien.
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.