Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: a mistake of Fruit2.2Uri in CEGT (bug or another problem?)

Author: Joachim Rang

Date: 13:49:49 10/19/05

Go up one level in this thread


On October 18, 2005 at 18:07:50, Uri Blass wrote:

>On October 18, 2005 at 17:44:15, Heinz van Kempen wrote:
>
>>Hi Uri,
>>
>>normally I like your postings a lot and the discussion with you, but this time
>>it was obvious that you were trying to find something with any means.
>>
>>So I hope that it is okay to finish it now and you can be happy that you found
>>something and maybe now scrutinize games from other testers. When I want to find
>>something I can do this with any bigger database.
>>
>>I am very sure that the games material offered by CEGT testers is above average
>>and valuable info for many authors.
>>
>>Subject finished for me by now.
>>
>>Best Regards
>>Heinz
>
>I tried to find something because I suspected something was wrong.
>
>Imagine that some tester get
>Djinn 0.892c-Fruit2.2 10-0
>I am sure that people are going to look at the games to try to find if something
>is wrong.
>
>It was not as extreme as the result that I suggest but it was enough for me to
>suspect that something is wrong.
>
>Note that the surprising results are not only of Fruit but also of Shredder
>Colombus that was supposed by sandro to be clearly better than the default and
>it seems now clear that it is not the case unless there is an hardware problem.
>
>Uri


The surprising part is that Sandro claimed it to be better than Shredder default
or to be more precise that you apparently believed in it as well that your
observation of a few postitions where Fruit did better with a threshhold of 50
lead to your assumption that a value of 50 could increase playing strength.

Normal expectation was that Sandro just tried something and cried Ura after some
statistical noise as well as that observing behaviour in a few test positions
does not tell something about playing strength. So CEGT data confirm the to be
expected results. :-)

Joachim

P.S.: I recommende testing threshold=50 since it is for us interesting to know
the behaviour of lower threshhold in longer games - I never expected them to
score better.



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.