Author: Uri Blass
Date: 05:46:14 06/05/02
Go up one level in this thread
On June 05, 2002 at 07:57:20, Gian-Carlo Pascutto wrote: >On June 05, 2002 at 07:11:27, Uri Blass wrote: > >>It is also the case in a lot of positions in the GCP test suite and I wonder if >>you think to have a different test suite in order to help you to test changes in >>the search rules. > >Hmm, in all positions in that testset, the difference between the moves >should grow to a pretty significant value given a long enough search. > >We did a run over all positions in K-best mode, and I excluded all >those were the difference was too small (tested in several programs). > >It's possible that for a particular program such positions still >exist, but there certainly shouldn't be 'a lot'. > >-- >GCP The problem is that there are positions that programs can find the right move when they do not see the tactics and there are some cases when movei changes it's opinion from the right move to the wrong moves or from the wrong move to the right move when the difference is only small. I suspect that if I give it more time it may see the tactics but in this case the test should be defined as not only to find the right move but also find at least some evaluation number. The question is when to consider the position solved. Not in all the positions the difference grows from the first moment and there are positions that movei changes it's mind from the right move to the wrong move and only later changes it's mind again and see the tactics. Uri
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.