Author: Uri Blass
Date: 12:48:37 12/12/01
Go up one level in this thread
On December 12, 2001 at 15:16:01, Brian Richardson wrote: >On December 12, 2001 at 14:25:34, Uri Blass wrote: > >>On December 12, 2001 at 13:43:17, Christophe Theron wrote: >> >>>On December 12, 2001 at 03:17:30, Uri Blass wrote: >>> >>>>My guess is that a piece square table program with good search rules can get >>>>2200-2300 ssdf rating on A1200 >>> >>> >>> >>>Much more than that. At least 2400 SSDF elo. >>> >>> >>> >>> Christophe >> >>It may be interesting if you can release also an engine with only piece square >>table evaluation. >> >>I am not talking about relative simple evaluation like the idea to check gerbil. >>Gerbil is not optimized for speed and it has not optimal search rules. >>Gerbil also does not use only piece square table evaluation. > >Gerbil version 0 or 1 was material and piece squares only, as I recall. >The source is still available, if you want to do some tests. >It is also a very fast program (in terms of raw NPS). No It is not very fast. Bruce told me that it is not optimized for speed and it means that it could be clearly faster with the same evaluation. I also looked at the source code of the first version and I found 2 tables for the king(I guess that one of them is for endgame) When I say only piece square table evaluation I meant only one table for every piece. >It hit about 2300 on a system about half the speed of an A1200. 2300 has no clear meaning when I have no idea about the time control and the rating system. I read in the release notes 2150-2350 against humans in 5 0. It is probably clearly weaker than 2200-2300 ssdf rating because it is known that computers are good at blitz. >Not sure there are "optimal search rules". What is good for a limited PCSQ >based evaluation may be poor with a richer evaluation funcion. I am almost sure that the search rules are not optimized because bruce did not care about possible small improvements. Uri
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.