Author: Dann Corbit
Date: 10:53:07 10/28/05
Go up one level in this thread
On October 28, 2005 at 09:44:07, David Dahlem wrote: >On October 27, 2005 at 13:57:57, Dann Corbit wrote: > >>On October 27, 2005 at 10:44:17, David Dahlem wrote: >> >>>On October 27, 2005 at 09:29:39, Jon Dart wrote: >>> >>>>On October 27, 2005 at 05:03:18, Thorsten Czub wrote: >>>> >>>>>the program i want to test has no EPD import but maybe it works >>>>>with FEN in PGN ?? >>>> >>>>Try Arena (http://www.playwitharena.com) - it can read EPD >>>>and drive the engine to analyze it. >>>> >>>>> >>>>>has anyone tough positions or knows how to get them into the prgram without >>>>>manually set up everything ? >>>> >>>>The positions I use are available at >>>>http://www.arasanchess.org/tests.zip >>>> >>>>--Jon >>> >>>Is there a test suite that will give an accurate and reliable estimate of engine >>>strength or improvement? My results on test suites has been very disappointing. >>>The version that scores best on suites always plays worse in actual games. :-) >> >>Test suites do not measure engine strength. They measure tactical ability. >>Hence, there is no sure correlation between test suite solving and game play. >> >>I would put it this way: >>Good tactical ability is a necessary, but not sufficient, condition for chess >>engine excellence. >> >>Test suites are good for at least two important things: >>1. Finding holes or errors in your evaluation >>2. Ensuring that you have not introduced a serious bug. >> >>The only way to find out if the program plays stronger is to play lots and lots >>of games against a diverse crowd of opponents. > >So, if i add a new evaluation term, and want to test and tweak the bonus/penalty >values for this new condition, or test if this new condition is better or worse >than the old version, test suites are useless for this purpose? They have some value. But don't expect that optimal for test suites is optimal for game play because usually it is not. It is good to find an approximate starting value and also perhaps a valid range of values to test.
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.