Author: martin fierz
Date: 23:41:23 01/21/02
Go up one level in this thread
On January 21, 2002 at 03:59:55, David Rasmussen wrote: >On January 20, 2002 at 19:21:07, martin fierz wrote: > >>On January 20, 2002 at 16:21:02, David Rasmussen wrote: >> >>> >>>I am not talking about things that depend on a lot of other stuff. Evaluation >>>should take care of this. I am talking about Crafty that explicitly checks for >>>some material special cases _and nothing else on the board_, and decides if it >>>should give a penalty/bonus. I say that I think it can be done with the values >>>alone. Why not? >>> >>>/David >> >>because all your examples above depend on a lot of other stuff maybe? that's the >>point i tried to make in my post: a single rook with pawn vs knight and bishop >>can have good chances in this pure endgame (+ a few more pawns each). in the >>middlegame nearly never. how would you want to encode that in your material >>values? > >I don't want to do that. I would want to add evaluation terms for that. Isn't >that necesary with 1,3,3,5,9? Yes it is. What I am suggesting is not catching >_all_ cases with material values, just to catch some more than one can with >1,3,3,5,9. And have fewer evaluation terms for special cases. I believe that is >possible. At least I haven't seen any arguments against it. It seems that most >people in this thread has misunderstood my question altogether. > i dont misunderstand your question. but there are lots of things you must take into account for the material evaluation, *regardless* of what your base values are. at least if you want to write a decent program. so in some sense, your question has no answer, because whatever your base values are, you should add some special cases. however, the reason that 1,3,3,5,9 is "magical" as you term it is that these are in fact very good values for the pieces. every beginner's chess book has these values in. i'm pretty sure that these values represent the values of the pieces as accurately as possible under the constraints pawn=1, values only integers. and i think these two conditions just come from the fact that when you write a chess book for beginners, you want it to be simple. if you were to ask me, i would rather go for 1,3,3.2,4.5,8.5 for PNBRQ, which i think is definitely better, but the real difference between players of different strength is not that they calculate such numbers well, but that they know certain rules about when a piece may be better or worse than it's material value (static things like bishops vs knights in open or closed positions, or pawns only on one wing, bishop pair, or things depending on other pieces being present, like: Q+N > Q+B, but R+N<R+B or B+B>N+B or an exchange down for a pawn being ok in the middle game but deadly in a pure R vs minor piece endgame). there are so many of these that you cannot catch them with the piece values anyway... cheers martin >>a single rook with a pawn against two bishops is mostly in deep >>trouble. >>you have to check these specific cases in the end anyway, so what is the problem >>with crafty's approach? just use normal values, and add the special cases. >> > >There isn't necesarily a problem. But if I can have fewer special cases, by >adjusting the material values which I will add anyway, I will. > >/David
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.