Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The truth about chess programs- 3000 Elo is very High, i am not sure

Author: Dann Corbit

Date: 11:47:24 04/26/05

Go up one level in this thread


On April 23, 2005 at 21:53:09, Tanya Deborah wrote:

>On April 22, 2005 at 12:59:20, Dann Corbit wrote:
>
>
>3000 Elo is very very High..  I really dont think so.
>
>If Hydra can play at this level in the match against GM Adams, It will be very
>difficult to Adams to has an advantage or a superior game in some of the six
>games. It means almost 300 elo points more. With that difference GM Adams only
>can draw, and this is very hard to believe. We will see.

Elo figures with a small number of games have very low precision.

Really, all we have for something like Hydra is a TPR sort of measurement.

It will take hundreds of games in tournament settings to develop an accurate Elo
figure.

The number probably does not mean what you think it does anyway.

A figure of 3000 Elo could be accurate.  But here is what the number would
represent:

1. Given the exact pool of opponents which were pitted against the machine
2. Given the exact playing conditions that were used in the matches

then a rating difference from the 3000 figure against those opponents and under
those conditions can be used to predict a broad result against a class of
oppoents from that list.

This value CANNOT be used to predict the outcome of a single match even with the
exact same opponents used in the calculations (except on a broad probability
basis).

This value CANNOT be used to predict the outcome against a pool of competition
against which the machine has not been tried.

We can extrapolate that a machine that did very well against (for instance)
Shredder would probably do very well against humans.  But it would be pure
guesswork.



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.