Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: This test is not scientific!

Author: Dann Corbit

Date: 11:59:21 01/26/99

Go up one level in this thread


On January 26, 1999 at 13:49:29, Don Dailey wrote:

>Hi Dann,
>
>I'm not trying to prove guilt, I'm trying to prove innocence.  I don't
>think this test can prove guilt but it can easily raise enough doubt
>that we should back off.
>
>If you are right and all or even one of the programs give a high match,
>then there is plenty of room for doubt and we give them the benefit of
>the doubt which I think is fair.   If however, Crafty matches
>significantly higher percentage, then we have something to talk about.
>I won't be claiming guilt in this case,  just that there is still
>room to talk.
>
>I have already discovered that running 60 seconds on Cilkchess will
>guarantee a much lower match rate,  so I am a little concerned that
>people will run this very short 60 second test and try to draw unfair
>conclusions from it.
>
>We will have enough data to come to some conclusion, even if the
>conclusion is that we are not sure!   We also have log files from
>them and an executable.  So we should be able to come to some
>intelligent conclusion.   If nothing else, I want to know for my
>own edification.
The point which I was hoping to make was that a 100% match does not prove guilt
nor does a 100% failure to match prove innocense.  This exercise can prove
neither (IMO - but I could be wrong, of course).

For example, I could change a couple numeric constants in the eval function of
Cilkchess and make it play wildly different.  It would probably take 15 minutes
of effort.  Or I could completely change the architecture and have it play much
the same.

My point is that the exercise does not have the intended effect.



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.