Author: Peter Berger
Date: 15:29:33 06/06/02
Go up one level in this thread
On June 06, 2002 at 17:37:54, Rolf Tueschen wrote: >I take as a basis between the two of us, that you can't get results for the >variable "chess strength" or performance _if_ you let vary (instead of keeping >em constant) variables like hardware or the program once without learning >function, because it was not yet invented, and with the function in modern >progs. These adds are gaps in computerchess. (When in human chess human beings >have differences too but _not_ in the absolute sense but relative, and these >diffs are distributed on a normal graph.) Your wording is still kind of unfortunate (from a maths point of view) but I think I get your idea here. First: I think you are basically right in fact. When the difference in strength between two chessplaying chess entities is too huge the assumption of normal distribution won't work well any more (as to be expected btw). But I think playing strength is the major factor here - hardware, opening book, learning, egtb, whatsover : shmanet. That's not what you would expect at first sight. It is against intuition. End of last year I played two matches ChessTiger on a PalmVx against weaker amateur engines on an Athlon1333 (that were expected to have similar strength from rating lists). The hardware difference was incredible. Basic difference was that in the first match Tiger wasn't able to ponder which had been implemented in the second version. From a hardware point of view that shouldn't matter at all - it is still an incredible difference. But the results seemed to follow expected distribution perfectly well, and the improvement seemed to be similar to what you would expect to get with programs of similar strength on same hardware IMHO. If you are interested I could search for the match results. Regards, Peter
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.