Author: Uri Blass
Date: 03:15:02 06/13/01
Go up one level in this thread
On June 13, 2001 at 05:57:11, Christophe Theron wrote: >On June 13, 2001 at 05:35:30, stuart taylor wrote: > >>On June 13, 2001 at 05:05:45, Marcus Kaestner wrote: >> >>> >>>> >>>>It is the most reliable tool that we have to evaluate the chess programs. The >>>>difference in the opponents each program has to face does not matter from a >>>>mathematical point of view. >>> >>>i think you are wrong! >>> >>>having now over two years of experience with my own chessbits-lists, i can say >>>to you that it is VERY easy to place a program with a HIGHLY overrated (or >>>underrated) position into the list. >>> >>>there are dozens of possibilities to fake a rating. >>> >>>i do not say that the ssdf fakes, i only say that it DOES matter which opponents >>>you choose. >>> >>>marcus >> >>Of course it does! >>If you play a 1200 elo rated player you can be sure to win 100% of times, and >>you play 100,000 games, you will have got a rating of 3000 long long before >>then. > > > >No you are definitely wrong. > >Due to the elo calculation formulas, a program which wins 100% of its games >against a 1200 elo opponent will have a rating of exactly 1600. > >Far from 3000, isn't it? > >Food for thoughts. It is truth by the linear formula but not by the right formula. The linear formula is only approximation to the right formula Here is a better formuala: If you score an average of P points in every game then your rating is 400logP/(1-P) more than your opponent when you use log by basis 10. It means that difference of 400 means that you beat your opponent 10-1 800 elo difference means that you beat your opponent 100-1 1200 elo difference means that you beat your opponent 1000-1 You do not get never 100% by this formula unless your rating is infinite. 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.