Author: Richard Pijl
Date: 01:50:27 08/31/04
Go up one level in this thread
On August 30, 2004 at 14:03:49, Dan Honeycutt wrote: >In spite of Dr. Hyatt's admonishment that this was a loser in Crafty, I'm bent >on trying it. Since I do legal move generation I can do the pins for half price >- I either have or can use the pin data for one side in conjunction with >generating moves. I just finished the routine. My NPS dropped by just over 4% >on bench test psitions. I've just started testing an hope to have some >indication if pin detection is a win, lose or draw in a few days. > >One aside - the pin detection is not always more accurate (or mine isn't). In >the following position for the move Nxg5 the no-pin SEE correctly returns +1 >where the pin-detection SEE comes back with -2. > >[D] 7k/4q3/8/6p1/4N3/5N2/8/4K3 w - - > >Fixing this would require keeping track if the pinning piece enters the >exchange, which I'm afraid would bee too expensive. I'm feeding both pinned and pinning pieces as a bitboard to my SEE routine. Whenever a pinning piece enters the exchange (easy check with bitboards) I'm examining whether the pinning piece releases a pin, or that another piece becomes the pinning piece. Of course, there are positions where this misevaluates as well, but generally it gives a better value. The penalty for taking pins into account in the SEE is not that big when you're already have those bitmasks available for evaluation purposes. I only use the pin-aware SEE in Qsearch as I don't have the pinned/pinning bitmasks available in inner nodes (something to try I guess, e.g. to use the pin/pinning data also in moveordering). Richard.
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.