Author: martin fierz
Date: 01:40:03 11/18/03
Go up one level in this thread
On November 17, 2003 at 23:56:43, Igor V. Korshunov wrote: >On November 17, 2003 at 07:28:28, martin fierz wrote: > >>On November 17, 2003 at 05:28:19, macaroni wrote: >> >>>What programs use this method for evuation? (perhaps varying it at the start of >>>each search). At the moment my program uses this, and a bit of king safty done >>>in-search. Is it a viable method? or before I re-write my fairly bad SetMaps() >>>function, should I change the whole system? >>>cheers >>>tor >> >>piece-square-tables have the advantage that you have a really fast evaluation >>thanks to incremental eval updating. > >I don't do PST incrementally. I have no complex evaluation. >An experiment: > >Speed of my engine: > - with PST - 749 kNp/s > - without PST - 753 kNp/s > - only material evaluation - 1083 kNp/s > >PST is very small part of simple evaluator. >Do you still think that incremental updates of PST is able to make your program >faster? well, of course the whole point is that PST is blazingly fast if done incrementally compared to any decent evaluation function. so if you use only PST and not much more, you can be very fast with it. given your numbers you won't gain anything with incremental updating of course, since your PST is a very small part of the evaluation. > i'm certain that a good >>evaluation will do better than PST. > >Of course, good evaluation is better than bad. But can you offer something to >replace PST? the whole PST thing seems to me to be a substitute for mobility analysis. so i suggest to get the mobility of your pieces instead of blindly assuming that a centralized piece is better. or for passed pawns really check whether they are blocked or dangerous instead of giving a fixed bonus for being on a certain square. or to check whether a piece is attacking squares close to the opponent king - etc, etc. of course this is much more expensive than PST! and given how weak my chess program is you probably shouldn't listen to me ;-) cheers martin
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.