Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Poll Question - Tournaments vs Matches

Author: Robert Hyatt

Date: 07:07:59 01/06/00

Go up one level in this thread


On January 06, 2000 at 09:02:51, Chris Carson wrote:

>On January 05, 2000 at 16:36:33, Robert Hyatt wrote:
>
>>On January 05, 2000 at 15:11:43, Chris Carson wrote:
>>
>>>Your opinion is valued.  :)
>>>
>>>Take a look at some of the tournament performance
>>>ratings (TPR) of some of the programs (Genius, Junior, Rebel,
>>>Tiger, Hiarcs, and others).  See the computer resource center,
>>>rebel web page, Gambit Soft tournaments page and U pitt.  Take
>>>a look a the AEGON and other events.  Also take a look at the
>>>ratings on ICC.  Top programs have high ratings there.  :)
>>>
>>>Be prepared to be shocked.  Sometimes the programs have high
>>>TPR's (above SSDF ratings).  Critics call these an aberation.
>>>Sometimes the programs have lower than SSDF ratings.  Critics
>>>call this the norm.  I think both are expected and normal,
>>>just as human ratings fluctuate within expected norms.  :)
>>>
>>>Best Regards,
>>>Chris Carson
>>
>>
>>This is bad statistical methodology however.  If you pick any program of your
>>choice from Aegon, I will pick one to offset it.  For any 2600TPR you find, I
>>will find one with a 2200 TPR.  That is the problem.  Statistics looks at the
>>averages, not at individual data points.  And the average is way below 2600.
>>
>>As far as ICC goes, Crafty has been over 3250 there.  I doubt it would have a
>>prayer in hell of doing that in FIDE events, running on a T932 even.
>
>Bob,
>
>You make some good points (as always).  :)
>
>My point was that a 2600 TPR can be found (and I am not surprised based on
>SSDF ratings given 95% or 99.7% confidence levels) for some programs.
>I agree that 2200 TPR's can also be found.  :)  I consider both ends of
>the spread valid for the programs associated with the numbers.  :)
>
>I think we agree.  :)
>
>Best Regards,
>Chris Carson


The point where we may not agree is on the question "Is the computer a GM yet?"

It is easy to find places where a computer produced 1 2600 TPR.  But to become
a GM, it has to produce 3, and all the while keep its rating over 2500.  That
is a lot harder than it is to produce one good result that is just a statistical
anomaly.

I think that the 3 norms would be very difficult to produce, and staying over
2500 while doing so (at least a year) would be even harder when everyone starts
to take notice.



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.