Author: Robert Hyatt
Date: 13:14:10 04/26/00
Go up one level in this thread
On April 26, 2000 at 16:10:36, blass uri wrote: >Some programs can have main line ends with a mate without understanding that it >is a mate(I saw that it happened to Crafty and Junior). > >I think that knowing if the position is checkmate or not checkmate is an >important knowledge. > >It is possible to see it by search but I think that seeing it by evaluation is >faster because you do not need to generate all the possible moves because >discovering one legal move is enough. > >I guess that the reason that programmers avoided to do it is that they did not >want their program to be slower and that they found that in most of the >cases(when there is no checkmate) they waste time without earning something for >it. It has to be done at _every_ leaf position. Which means it suddenly becomes a time-critical thing to do that is very expensive. I don't notice it causing any serious problems as eventually _every_ program has to stop the search. In Cray Blitz, I didn't stop if I was in check. But the hardware made the test almost free. And when I later compared crafty to CB in lots of positions, I never saw this as a problem. CB saw more, but it had better extensions. But it never saw something that crafty didn't, just because crafty didn't notice one side was checkmated... > >How much time does a chess program need to find if a position is checkmate or >not checkmate? > >Uri Not a lot of time. But the problem is that a big number (number of leaf positions) times a small number (time to do the checkmate test) turns into a big number, and slows the search significantly. Question is, does the slow-down pay for itself or not. I used to think so, but after 5 years of testing, I don't fool with it now.
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.