Author: Robert Hyatt
Date: 07:41:18 01/24/99
Go up one level in this thread
On January 24, 1999 at 06:51:45, Steve Maughan wrote: >I understand the basic idea behind SEE is to evaluate the material outcome of a >position without searching - however I have some questions > >1 What algorithms are used to do this? It looks none-trivial since one cannot >look at a single piece and decide whether or not it will be captured as you may >also have a more valuable piece en-prise. > you enumerate all the pieces for white and black that are attacking the single square you are applying SEE to. Then you play thru the captures always capturing with the smallest piece possible (and when you capture diagonally with a piece you _then_ look behind that piece to see if another piece is now attacking since that piece moved out of the way, ditto for files, but not for knights) Once you play out the captures, you minimax the scores to see when one side would decide to stop (ie white might play PxN, black plays NxP, and white is +2 ahead here. If the only capture left for white on that square is QxN, but black can then play BxQ, then black would be winning. So the minimax algorithm decides who will refuse to continue the captures and take the score from that point... _all_ of this done on a single square. >2 With 1 in mind - do you believe the SEE outcome enough to back the score up to >root or simply as an indication of moves to search further? I don't. I use it for move ordering. Or in the q-search, I use this to say 'no need to even search this move (ie QxP is a valid capture, but no point in searching it if the opponent replies NxQ). I don't use it for evaluation, only for ordering captures in the tree, and excluding them in the q-search. > >3 Are there any alternatives? Does 'everyone' use SEE? Another ordering strategy is MVV/LVA (most valuable victim, least valuable attacker). Here you generate captures, and try to capture the most valuable piece you can with the least valuable attacker you have. This doesn't do so well when your opponent has an unprotected pawn, but you first try QxR (you have to first capture the most valuable piece that is attacked, using your least valuable piece to capture). You'd rather try QxP first as the pawn is 'free' while QxR probably loses a Queen for a rook... Again, only for ordering the capture moves, of course... > >All help greatly appreciated! > >Regards > >Steve Maughan
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.