Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: how not to calculate performance

Author: Sune Fischer

Date: 10:24:00 10/29/04

Go up one level in this thread


On October 29, 2004 at 12:27:15, Uri Blass wrote:

>Let assume that every player play 12 games against 1400 players
>I want to give an estimate for the level of the player based on the result after
>4 games.
>
>part of the players who score 0 out of 4 will score more than 0 out of 12
>
>A good estimate remain the same in average when you play more games(it becomes
>higher for players who do more than expected but also becomes smaller for
>players who do less than expected).
>
>If your estimate based on 0/4 is the same as your estimate in case of 0/12 it
>means that players who score 0/4 can only improve their rating after they play
>more games and it is not logical.
>
>The average level of players who score 0/12 is lower than the average level of
>players who scored 0/4 because players who scored 0/4 consist also players who
>can expect to score results like 1/12.
>
>Uri

A players rating is connected to his expected score.

I'd say if he scores 0/4 the best guess we can make on his expected score
is 0.25/4, so right between getting nothing and getting a single draw.

This is of course because the actual score is descrete while the probability
is continuous, he cannot score his expected in that short a match.

So assuming he scored 0.25 of 4 against a 1400 player his estimated rating
(big error margins of course) would be 930.

If he got 0 of 12, his expected score 0.25/12 would give a rating of 731 ELO.

Likewise, had he won 4/4 (ie. 3.75/4) he would be rated 1870 and 12/12 would
give 2069 ELO.
If he wins 100/100 the rating would be 2440, 1000/1000 it would be 2840.

So I guess a 1400 player would be expected to draw Kasparov once in abourt 1000
games.

-S.



This page took 0.01 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.