Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SSDF Rating list

Author: Ulrich Tuerke

Date: 06:36:59 06/15/01

Go up one level in this thread


On June 13, 2001 at 14:25:10, Robert Hyatt wrote:

>On June 13, 2001 at 07:04:07, Ulrich Tuerke wrote:
>
>>
>>The opposite is true. When 2 payers with a huge difference in ELO play, the
>>stronger player can only loose ELO points. In case he gets 100%, his rating
>>remains constant, otherwise it goes down.
>>
>>Uli
>>
>
>
>That is wrong.  Either (a) you are using a TPR rating estimator like
>
>new = old + diff*400*(win-lose)/N
>
>That does limit you to 400 points difference, max.
>
>or (b) you are looking at the chess servers, which use integer math to
>compute ratings.  That also gives you a bounded upper rating as soon as
>the rate of change drops below 1.0 it becomes 0.
>
>But Elo's formula has no such limit.  I can be rated 4000 and you can be rated
>1, and I will _still_ see my rating go up a small fraction of a full point.

Oh, I didn't know. Thanks for clarification.

Nevertheless, the original claim that a program can easily get a rating of 3000
by playing only weak opponents, is wrong. One would have to gather endlessly
fractions of an ELO point (and finally running the risk to spoil everything by
having just one drawn game).

Uli



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.