Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Looking for clean test suites any ideas.

Author: Shaun Brewer

Date: 16:57:42 03/04/06

Go up one level in this thread


On March 04, 2006 at 18:20:12, Charles Roberson wrote:

>
>>> At this rate it will be ages before I even have a trusted suite.
>
>>>Shaun
>
>   Yes, that is the problem.
>
>   So, I suggest doing the correlation tests with several suites and allow
>   the engines say 2 minutes at best per position. (this could take months --
>   I know, it took 4 weeks for me and I divided it out amongst 4 computers).
>
>    At that point, you will be able to tell which suite is the most correlated
>with strength. Then, make that suite perfect.
>
>    If you start with making a suite perfect, you may spend a month on it and it
>doesn't correlate with performance. Now, you start over with another, then
>another .....

That is probably a good idea - check if the suite gives a result similar to game
performance and then check if all the bm am etc are correct.

So far I have been testing the MATS suite:

http://www.cs.unimaas.nl/~uiterwyk/icca/mats.htm

and

the suites on Alessandro Scotti's home page author of KIWI

http://www.ascotti.org/programming/chess/chess.htm

I have just realised searching for tets suites at home rather than at work in my
lunch break is much more profitable as dozens of sites are blocked at work.

(including this one)

I will keep you posted

So far I seem to be proving that poor test suite = good game performance.

MATS (15 minutes on my Laptop)

GFruit-4bx   18/24
TogaII 1.1a  16/24
Spike 1.1    15/24
Glaurung     15/24
Rybka        14/24!!!

From various rating lists it would see I should hope for:

Rybka
Spike-Toga-GFruit
Glaurung

Anyway I will continue to search

Shaun




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.