Author: Jon Dart
Date: 09:11:07 01/20/03
Go up one level in this thread
On January 20, 2003 at 00:15:31, Scott Gasch wrote: >Hi all, > >I'm pondering a major code overhaul of the engine to implement some ideas I've >been thinking about for a while. One of these is attack tables. So I wanted to >seek the advice of those of you who have attack tables in your engines already. > >First, I assume attack tables are things you want to do incrementally. This >seems really expensive though Arasan used to have incrementally updated attack information. If I remember right, it was lazy about it and didn't deal with stacked attackers, so the attack info was not really useful for SEE. Now it uses rotated bitboards and can do an accurate SEE. Generally, I now think that you want to do as little as possible in your "make move" routine. The faster your h/w is, and the deeper you go, the more cost you are paying if you are loading things into "MakeMove" and "UndoMove". Of course doing this can make your eval faster, but then some of the nodes you are moving to won't even be evaluated because of pruning, etc. Overall incremental attack generation cost Arasan speed, and taking it out was an improvement. Of course, your mileage may vary. My overall impression from reading Ed's stuff is that many of his techniques work well together in his program but are not readily usable in a more conventional program. Rebel has traded speed for reduced tree size. It is a very interesting approach and evidently works well, but most programs are tuned to go as fast as possible. --Jon
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.