Author: Gian-Carlo Pascutto
Date: 00:23:49 06/18/02
Go up one level in this thread
On June 17, 2002 at 20:01:39, Russell Reagan wrote: >Secondly, do you think either approach is superior to the other? I think the >"natural" method would seem to be better. For example, if you reach an endgame >and castling still happened to be legal, and your program castles to get rid of >the "non-castling" penalty, even though your king should be trying to get >centrally located in the middle of the board. If your program had "king safety" >factors instead of the castling work around, then it wouldn't castle and would >play a more correct move. Seems pretty straight forward to me, but I've been >wrong before. I agree here, in that I tend to make the evaluation as 'natural' as possible. I do have exceptions though. i.e. my kingsafety consists of the generic code, followed by some hard-coded exception patterns. It might be possible to write a better generic function, but the current method works, and was easier to program. >Other examples could be using a piece-square table with low values for the >border squares and high values for the central squares so that the program >"controls the center" in the opening. Hmm, I consider centralizing the pieces to be a 'natural' rule. >Or penalizing pieces that haven't moved >yet in the opening, or penalizing pieces that have moved more than once, etc. I do this as well. It would be better if the program avoided this via more general rules though. In fact, I need to test whether they are still needed. My program doesn't really excel at understanding board control, so probably it does. >These seem to be very poor methods of approaching an evaluation function at >first thought, but I haven't worked extensively on my own yet, so I'm no expert >here. > >I'd like to know your thoughts... I very much dislike any kind of assymetric evaluation for the same reasons. Also things like fixed game stages. Not to mention preprocessing! Urgh! -- GCP
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.