Author: Ron Murawski
Date: 08:54:43 11/15/02
Go up one level in this thread
On November 15, 2002 at 04:20:56, Daniel Clausen wrote: >On November 14, 2002 at 13:34:05, Ron Murawski wrote: > >>On November 14, 2002 at 03:26:47, Daniel Clausen wrote: >> >>>On November 14, 2002 at 01:47:38, Ron Murawski wrote: >>> >>>[snip] >>> >>>>A computer never knowingly plays a true sacrifice. All it can do is make the >>>>move that will get it the highest score, aka "best move". >>> >>>And that is different to how humans play a sacrifice exactly how? Computers >>>are a bit more number-centric than humans, but that is true for the >>>non-sacrifice-moves as well as the sacrifice-moves. >>> >> >>The distiction is this: a human can make a move based on gut-instict or based >>on experience from playing other similar positions, a computer cannot. > >My point was that this is also true for non-sacrifices. > >And about the gut-instinct: there was this Novag board computer several years >ago (ForteC or something was its name, but I'm not sure. maybe Guetti knows? :), >which played so-called PSH-moves (in german: passt sicher halbwegs) which were >moves like Bxh7+ in situations where the computer thought that this could be a >good attacking chance, although it didn't see the full consequences in the >calculated tree yet. > I'm trying to do that in my own engine. > >>If a chess player knows his opponent he might play a slighly inferior move >>knowing that the other player is uncomfortable in certain situations whereas a >>chess engine will never play a slightly inferior move. > >Things like "asymetric king evaluation" could do that too. (if it's turned of >against GMs but not against 'normal humans') > > >>In order to get a computer to play a true sacrifice, you have to give a large >>enough positional bonus to fool the engine into thinking it's gaining >>something. > >Define "true sacrifice"? > The computer loses material and cannot win it back in the lookahead. The material piece gets traded for purely positional evaluation points. >It's a bit difficult to discuss this stuff.. What am I supposed to tell you when >you tell me that the Kramniks play sacrifices because of intuition, gut instinct >and whatnot, and if a computer plays a sacrifice then it's not a "true >sacrifice" because it's just some positional bonus somewhere.. > I agree, this is a very difficult subject just to define. >What if I'd do the following in my engine: >For every "sacrifice" (for the moment, I define a move a sacrifice if the engine >doesn't see a material equality/advantage after the PV - hey, at least I come up >with a definition! ;) my engine plays, I mark this move (eg Bxh7+) in a >permanent hashtable/list. Now my engine plays in a tournament and sees 2 moves >as promising in a certain position: > >Be4-e2, with a score of +0.5 (material +0, eval +0.5) >Be4xh7+, with a score of +0.4 (material -2, eval +2.4) > >Now my program looks in this other 'sacrifice-list' I mentioned before and >decides to play Bxh7+ because it won 10 games with this sacrifice before (and >only lost 1), whereas the move 'Be2' doesn't have an entry. > >What is this now? If a human does that, it's probably "based on experience". If >my engine does that it's "just adding some positional bonuses"? > That would be a way of imitating human behavior. Interesting idea... >Of course it would be a clever list, which would only add "Bxh7+" to the list, >if blacks kingside would still be in good shape otherwise (pawns on f7, g7, h7, >rook on f8, king on g8) And other things which improve the 'quality' of the >sacrifice-list. :) > >Sargon Sorry, but I'm about to leave for a week or so. Not much time... Ron
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.