Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Next Human vs Computer ratings list - I need opinions

Author: Robert Hyatt

Date: 10:19:37 05/19/00

Go up one level in this thread


On May 19, 2000 at 10:38:18, Enrique Irazoqui wrote:

>On May 19, 2000 at 10:30:26, Keith Conary wrote:
>
>>On May 19, 2000 at 09:42:07, Enrique Irazoqui wrote:
>>
>>>On May 19, 2000 at 09:37:19, Chris Carson wrote:
>>>
>> Hardware failures and operator's errors are part of
>>>how a program plays. Forfeits and IP failures are not.
>>>
>>>Enrique
>>>
>>I disagree.  Hardware failures and even operator errors have no bearing on the
>>performance of the programs algorithm.  If a true indication of performance
>>between Human and programs is to be revealed here, games where the program
>>actively causes a loss due to calculations should be included with the exception
>>of the Forfeits and IP failures.
>
>As I said in my reply to Uri, I think that games lost due to operator errors
>shouldn't count. But hardware failures are a different matter, in my opinion.
>You can discard a game lost because the machine was overclocked and failed, but
>then you should also drop all games played on that same machine, maybe won
>because of the overclocking. In any case, this might be a theoretical issue
>only, since I am not aware of any games counted by Chris that were lost for this
>reason.
>
>Enrique
>
>>Keith


I disagree with your disagreeing with the original disagreement.

:)

A computer entry in a tournament has several parts:  (1) program (2) hardware
(3) operator.  A problem by _any_ of the three components is going to affect
the game result at times.  And since the probability of error by any of the
three parts is non-zero, _all_ of the possible errors have to be accounted for.

IE if you enter program X in an event, and in one of 10 games a hardware glitch
causes it to lose, do you not count those losses?  If in one of 20 games a human
error causes a loss, do you not count those?  Ie the losses _did_ occur.  Just
like humans make tactical errors although not in every game to the same degree.

Is the purpose of this rating calculation to show how a program would _really_
do in competition with human GMs?  Or is it to show how a program would
_theoretically_ do if no hardware/operator errors ever occurred?  I think that
the former is the right answer, not the latter.  This is an experimental
setting, not a theoretical one...

Data obtained from the experiment must be used in its entirety...
With the one exception of games not played, which should not be counted since
they never happened...



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.