Author: Dave Gomboc
Date: 22:37:54 07/23/99
Go up one level in this thread
On July 23, 1999 at 12:43:13, David Eppstein wrote: >On July 23, 1999 at 09:10:37, Vincent Diepeveen wrote: >> I don't doubt that at the icc server using last ply pruning >> and not using checks in qsearch work together very well. >> >> In diep i don't use last ply pruning yet (although i experimented >> with it of course a lot) and therefore i can do checks in >> qsearch. > >My concern with not doing checks in the qsearch is that you will misevaluate >positions in which something is protected indirectly by tactics rather than >directly by another piece. The q search is supposed to resolve the current >material balance, so it should look at checks that change the material balance: >skewers, forks, and discoveries. > >I am not expecting qsearch to find mating attacks, that's what the regular >search is for and that way lies huge blowups in the qsearch tree size. If you >restrict yourself to only checks that affect material, I think the tree size may >not blow up so much. And, I think, one can use SEE-like techniques to >distinguish those checks from the other ones. > >But in Crafty's case, it is hard to argue with success... If the goal is to allow some, but not all checks in the quiescence search, what criteria will we use to decide which are permitted? Here are some possibilities, thrown out just to start the ball rolling. I have never tried any of them, and some might be very bad. Follow a check further when ... 1. ... a check allows only a single response. 2. ... another attack is also created when the checking move is played. 3. ... defensive square control around the king suggests that the opponent's pieces are poorly placed, while the player doing the check has a good amount of firepower to bring to bear on the attack. 4. ... it forces the opponent to give up castling privileges. 5. ... our own king is in danger (possibly trying for perpetual, but really to relocate our piece for defense of our own king.) I dunno, but that's a start. Some of these may be easier or more difficult to calculate, depending on how much the engine knows about the position its searching. Dave
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.