Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Puzzled about testsuites

Author: Michel Langeveld

Date: 09:02:47 03/09/04

Go up one level in this thread


On March 09, 2004 at 11:43:51, Tord Romstad wrote:

>On March 09, 2004 at 11:28:16, Michel Langeveld wrote:
>
>>I am very puzzled. Somewhere I should have made a mistake. Maybe my approach for
>>tuning on ecmgcp.epd was wrong.

>Yes, tuning for *any* finite number of test positions is wrong.  Playing a big
>number of games against a wide variety of opponents is the way to go.  It is
>painful, time-consuming and boring, but so far I haven't found any other way
>of testing which works.

Hi Tord!

Thanks for answering

I am affraid you are right ... I test most of the time a lot and make tiny
steps. What timecontrol do you use for testing games? I think I need also more
modern . The quality of faster hardware is automatically better than on my
hardware on the same timecontrols.

>>My questions:
>>* Is 5 seconds a move a good timecontrol to test ecmgcp.epd ?
>
>You should use more, IMHO, especially considering your slow hardware.  I do
>most of my testing on a PIV 2.4 GHz, and I never use less than 10
>seconds/position when running my engine through ecmgcp.  Version 0.4.6
>solves 136/183 at 10 seconds.

I think 10 seconds on your hardware is about 20 seconds on my hardware.
I will try to use a longer timecontrol for ecmgcp.epd too to see what happens.

>>* Does it makes sense to use a reall long timecontrol?
>>* I learnt that ecmgcp.epd is much more tactical as strategical.
>
>Definitely.  It is a collection of combinations, nothing more.  A purely
>tactical test suite.
>
>>Does someone
>>has other/ better positional/eval-tweaking advice? No testsuites? Other
>>testsuite, like lct2.epd, the comeplete ecm.epd ?
>
>The complete ecm.epd would be no better.  All positions are tactical.
>LCT2 is great, but unfortunately I don't know any other suites of
>similar type and quality.

What timecontrol do you normally use to test this? Do you look also for each
position to number of nodes searched. ply depth solution found?




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.