Author: blass uri
Date: 15:19:07 05/12/00
Go up one level in this thread
On May 12, 2000 at 17:58:35, Mike S. wrote: >On May 12, 2000 at 16:21:01, blass uri wrote: > >>(...) >>Unfortunately tactical test positions are usually sacrifices. >>(...) >>I think it is a good idea to generate a tactical test suite based on positions >>from practical games and not based on finding sacrifices. > >I have some experience with program testing. There's a simple reason why most >tactical tests are sacrifices: You need an unusual solution move, which would >not be played unless the program sees the special reason, in other words, it has >calculated the complete combination. >When the solution is a silent, and quite normal move or maneuvre, this is much >more difficult to judge upon. You would have to acquire an expertise for every >single test position (and for each program tested), if it hasn't only played the >awaited move, but if it has also really seen the reason and would play the >correct continuation. The test that I suggest is not about finding the right move but about finding a clear change in the evaluation(in some cases there are more than one right move). You can decide that if the evaluation is changed relative to the evaluations at small depthes by at least 0.7 pawn then the program solved the position. I think that this test is needed because it is possible that a change in the program will help it to find sacrifices faster(for example by reducing the value of pieces) but will not help programs to find tactics faster. 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.