Author: Martin Schubert
Date: 07:28:53 11/11/01
Go up one level in this thread
On November 10, 2001 at 11:22:24, Christophe Theron wrote: >On November 10, 2001 at 10:55:52, Thomas Mayer wrote: > >>Hi Thorsten, >> >>>ok - you come with games : very fast PC, more >>>than 83 times faster, and want to estimate a rating. >> >>>i do it different. >> >>>we have then vancouver 68020 and the diamond 2 and very precise >>>ratings of them. >> >>>as far as i have played out the GAMBIT version of tiger14.6 for palm >>>beats them. >> >>well, but let the vancouver play against some engines on Athlon 1333. What do >>you think will happen ? Do you think the vancouver will get the rating the >>rating that it has on the SSDF list ? I doubt that. > > > >Let me repeat it for the third time, and maybe you'll get it this time: > >by letting very different players (in term of strength) play each other YOU GET >A RATING WITH A HUGE ERROR BAR. It's not only a problem of a huge error bar. When you make two different lists (for example one for humans and one for humans) you can't compare the strength of one person from one list with a person from the other list. Even if you play thousands of games. Regards, Martin > >For example if you let Chess Tiger for Palm play Athlon 1333 top programs only, >you are probably going to get an estimated SSDF elo of 1000 for Tiger, PLUS OR >MINUS 1500 points. > >It's not about fairness. It's not about rating inflation. It's not about being >out of sync. It's about mathematical accuracy. > >These matches are useless not because they are not fair. > >They are useless because after playing them you have collected almost no >information, mathematically speaking. > >On the other hand I agree with you on the fact that subtracting 100 elo points >to the bottom of the SSDF list was unfair, because accurate ratings had already >been established for those slow computers. > >But this has nothing to do with the first point. > > > > Christophe > > > > >>So the main point is: Rating lists with big hardware difference get extremely >>out of sync... As I have said, I don't see any reason why e.g. the vancouver >>get's weaker in the last some years ? Do you see any ? By the way, many of those >>old dedicated chess computers have proven their rating against humans, so it was >>a little unfair to them to reduce their rating by 100 points... As long as >>anyone claims that SSDF shows ratings compareable to human ratings... Agreed ? >> >>Greets, Thomas
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.