Author: Colin Frayn
Date: 07:09:41 03/12/00
Go up one level in this thread
On March 11, 2000 at 15:03:58, John Coffey wrote: >I wonder how much benefit there would be to doing a dynamic exchange evaluation >on certain moves and just eliminating them from the search. Remove moves that >seem to obviously lose material. We might allow sacrifices at the base of the >tree and sacrifices that give check at any point. If you could just cut out moves which you played then that would be great, but overlooking good sacrifices that the opponent could make might seriously reduce your program's strength. Remember that they will mainly be playing other programs, and other programs won't be so prejudiced against sacrifices as humans are. By that I don't mean that humans hate playing sacrifices, just that they are naturally more prone to discounting them and missing purely tactical lines. A computer will easily spot a clever checkmate which is initiated by a sacrifice without any problem at all, whereas a human often has great difficulty in doing so. As an example, study the 'immortal game' (anyone have a reference?). A computer would never play Rg1 sacrificing its bishop anyway, so there's no point arguing about that move, but the method you described would totally miss the CM starting with Nxg7+ and then the queen sac. That's a very easy mate in 3, and ColChess spots it in less that a fifth of a second. You've got to expect that other (better) programs will spot things like that even easier (and deeper). >The benefit is that it would cut down on the size of the tree, and perhaps by a >great deal. I don't think so. Sure that's correct (in some cases) with a very naive search, but once you've added in lots of the better pruning algorithms and a hash table, I don't think that the speed up would be that great at all. Basically, what you've described sounds like a very strong variant of futility pruning, which basically stops the search when it's near depth of 1 or 0, the line you're following seems to be losing a lot of material, and it doesn't look like you're going to get it back in the near future. I've always been sceptical of this method, and I think Bob Hyatt dislikes it too (from the comments that come with Crafty's source code) - but note that I think Crafty uses razoring, which is similar but less foolhardy. >The cost is that the program would be blind to many tactical shots. >It would probably miss that a piece is pinned and cannot recapture. Why is that? Bottom line : bad idea Cheers, Col
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.