Author: Robert Hyatt
Date: 07:21:12 08/19/03
Go up one level in this thread
On August 19, 2003 at 05:04:58, Omid David Tabibi wrote: >On August 19, 2003 at 04:12:45, Tord Romstad wrote: > >>On August 19, 2003 at 02:48:36, Omid David Tabibi wrote: >> >>>It seems that some programs use Static Exchange Evaluation in order to prune >>>losing captures in quiescence search. >> >>Not just some. Almost all strong programs do this, I think. >> >>>In the following position, an SEE will deem the move 1.Rxd7 a losing capture, and it might >>>get pruned in quiescence. However, in fact this is a winning capture since Qf6 is attacked >>>after 1.Rxd7. How do such programs solve these kind of problems? >> >>They don't (or at least most don't). On the other hand, the significant extra >>speed gained >>by not searching all captures help them find a lot of other tactics which they >>otherwise >>wouldn't have found. >> >>Try it yourself. I am almost certain that you will find that excluding losing >>captures >>from the qsearch is a big win. > >But a good SEE will be quite costly by itself. For example: > >[D]3r2k1/pp1r1qpp/2pb4/5p2/3R1P2/8/PPPR2PP/3Q2K1 w - - 0 1 > >My engine has information about all attacked squares, e.g., it knows that d6 is >attacked by a white rook and defended by a black rook. However, I assume that an >SEE should be able to find out that Rxd6 is a winning capture. This needs a more >extensive processing of backed-up attacks (e.g., Rd4 is backed up by Rd2 which >is backed up by Qd1, and Rd7 is backed up by Rd8). I'm afraid this is too costly >a process. Not necessarily. Crafty has _always_ handled this case with no problems. As did Cray Blitz. As does most other programs I'll bet. When you do the capture analysis, and you "use" a piece, you just look behind that piece (only in one direction) to see if there is another piece that is not attacking the target. The expense is trivial. > >> >>Tord
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.