Author: Djordje Vidanovic
Date: 16:16:52 10/20/98
Go up one level in this thread
On October 20, 1998 at 12:56:29, Ulrich Tuerke wrote: >On October 20, 1998 at 12:29:59, Bruce Moreland wrote: >> >>60% is 100 Elo points, > >If I'm not wrong, 60% is rather in the range of a 50 - 60 ELO points >difference. But it doesn't matter anyway, see below. > >>if you can assume that 60% is the "true" result for these >>programs, which you can't from this sample, I would bet. > >No doubt about that. A sample of 200 games still has an uncertainty of about >50 ELO points. So, a 20 game match has an uncertainty of several >hundred ELO points. > >IMHO, the only rather safe conclusion is, that none of both programs is >completely without chance against the other one. Not a very sophisticated >conclusion, I'm afraid. > >Uli >>bruce The books say that 60% is actually a 72-rating-point difference, but never mind. What is at issue here is the small sample, so that both B. Moreland and Uli Tuerke are basically right. Let me just add that I consider the whole matter of rating as a daft little game made up of mere pointers, indicators of a sort, that present us only with a very general idea about a program's strength. Thus we can only know whether a program is strong, less strong or a duffer, but not really whether it's rated 2400, 2476 or 2512... All the figures are silly by themselves. And what is even more a laugh is that so many programs have been purchased due to their being rated one or two ELO points higher than others... Regards, Djordje
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.