Author: Bas Hamstra
Date: 15:12:57 11/24/02
Go up one level in this thread
On November 24, 2002 at 11:10:17, Robert Hyatt wrote: >On November 24, 2002 at 10:08:06, Ulrich Tuerke wrote: > >>On November 23, 2002 at 00:57:57, Anthony Cozzie wrote: >> >>>The discussion on verified null move pruning has got me thinking about my >>>qsearch. Right now zappa does simple-stupid qsearch: captures and recaptures, >>>and I'm thinking about adding some things. If I have a better q-search it may >>>be possible to move to straight R=3, which would be faster than the current >>>R=2/3 (Heinz). So I am considering adding: >>> >>>1. Pawn Promotions. It would be very easy to add some sort of "generate white >>>pawn promotions". Usually there would be none, but this could be determined >>>quickly by a bitboard and (for those of us smart enough to use bitboards . . . . >>>hehehe) >>> >>>2. Check evasion. Right now if one of the captures puts the King in check in >>>qsearch, zappa just stops. >>> >>>3. A Horizon zone. For example, Search() would call horizon() which would call >>>QSearch(). In the horizon zone, Zappa would also search checks and killer >>>moves. >> >>How about >>4. Check generation. For example, you could generate checking moves in >>quiescence provided some conditions are fullfilled (e.g. close enough to >>horizon, checked king standing risky, ...). >> >> >>> >>>Secondly, what are people's opinions on SEE versus futility pruning in QSearch? >>>I was using SEE up until recently, when I did some experiments and decided >>>Qsearch was working better. Of course, I may just have bugs in my SEE. >>> >>>I'll probably try most or all of this stuff anyway, but I'm curious what other >>>people's views on this are. >>> >>>anthony > > >I've mentioned this before, but in Cray Blitz, and some very early crafty >versions, I had a three-zone search, rather than the current two-zone search. > >Zone one was normal. > >Zone two was selective and included mainly tactical moves such as checks, >captures, moves that contain some sort of threat (found by null-move search) >and so forth. Interesting. How do you find tactical moves by a nullmove search? Isn't that very costly? How did you do it exactly? It seems a bit extreme to nullmove and do a 1 ply search, just to find a good move in a qsearch? Best regards, Bas.
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.