Author: pete
Date: 12:55:11 07/10/00
Go up one level in this thread
On July 10, 2000 at 14:56:31, Christophe Theron wrote: >On July 10, 2000 at 14:15:39, Terry Ripple wrote: > >>I know Hiarcs7.32 is one of the most knowledge based programs, but what about >>the famous Shredder4, Rebel Century and Junior6? Where do they average on >>knowledge in comparison? >> >>Best regards, >>terry > > >For me the answer is easy. I know others will disagree... > >The program that knows the most about chess, and has the most relevant >knowledge, is the one that stands the highest in the rating lists. > > > Christophe Sure this would be true if there were such things as trustable rating lists . a.) Swedish list ? Judges strength in comp-comp ; nobody of the people who believe in the swedish list will object to this obvious statement b.) ICC ? Completely untrustable ; everyone who has run a computer account there will know ; you measure various things ( like operator behaviour , picky or take anybody ) but only to some very mediocre extent find out about the performance of the program ( and probably _everyone_ running a prog account on ICC will agree ) . ICC results without close look at the played games and the general rating trends are simply useless . But ICC helps you to notice the significant differences between comp-comp games and comp-human and a "true" ratinglist would include games against both groups . For example changing hardware with a quite liberal formula produces funny results ; you will most likely notice that results against humans nearly don't change at all ( you lose against the same guys and win against the same guys ) . But the results against computers will _considerably_ change if your program runs relatively bugfree. I have done this experiment with a PII300 and a PIII500 multiple times and am quite sure about it . There is a really interesting post by Ed Schroeder in the Rebel Forum where he discusses differences between Century 1.0 and Century 2.0 which were about speed improvements and rating influences on comp-comp and conp-human and I completely agree . I haven't written a chessengine but this effect is so obvious and beyond any statistical doubt IMHO that I am surprised it is not simply widely accepted . So if this is true , "chessknowledge" ( either adding _or_ removing ) obviously will be the thing having more influence against humans . And then it is logical to me that the Swedish List or any comp-comp list doesn't help to judge about the _knowledge_ of chessengines at all when it is about games against humans. pete
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.