Author: Robert Hyatt
Date: 20:31:59 01/04/01
Go up one level in this thread
On January 04, 2001 at 16:29:17, Severi Salminen wrote: >On January 04, 2001 at 10:03:11, Robert Hyatt wrote: > >>On January 04, 2001 at 04:05:02, Severi Salminen wrote: >> >>>Hi! >>> >>>I'm planning on programming SEE into my program today. I'd like to know whether >>>I "should" (well, of course I can do anything I want but) prune all losing >>>captures in qsearch or search them (after the good ones)? And what about equal >>>captures? What I mean is that if I prune, I get a speedup, but I lose a lot >>>positional info if that particular losing capture leads to (say) passed pawn or >>>similar. What do you do? >>> >>>Severi >> >> >>I do something close to that. I take the current material score, plus the >>'score' computed by SEE, plus a fudge factor, and ask "does this at least >>get close to or more than alpha?" If yes, I search it. IE if I am a queen >>down, then winning a pawn won't help and it is futile to search the move >>even though SEE says it wins a pawn. > >So this is like doing a more accurate futility pruning in qsearch. Do you use >the same margin as in futility pruning (I use 2 pawns). If I have understood >right, futility pruning in normal search does only need to know the value of >piece being captured, not the true swapoff value so SEE doesn't help there. > >Severi I think I use one pawn. But my original "estimate" is not really just the material score, it is a true call to Evaluate() since this has to be done to fill in the "stand pat" score anyway. It is very unlikely a single capture will change the material score in a huge way except when the last piece is removed, etc... and you can factor that in.. You can use just the value of the piece being captured for futility pruning, but it is more accurate to use the SEE score instead, as it is a better estimate of what will happen...
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.