Author: Russell Reagan
Date: 10:30:04 12/27/03
Go up one level in this thread
On December 25, 2003 at 20:32:24, Christophe Theron wrote: >I have >already stated myself several times that simple evaluation terms are able, given >enough depth, to understand more complex concepts. So search is able to extract >some positional information that is not explicitely described in the evaluation >function, yes. At first this didn't make sense to me, but let me see if I get what you're saying. Let's say that there are two kinds of chess programs. Static evaluators and dynamic evalutors (both still use static evaluation functions). A static evaluator will evaluate things using static scores. Piece square tables, assigning a penalty for not castling, a static penalty for a isolated pawn, and so on, are examples of things that a static evaluator might do. A dynamic evaluator would compute things on the fly. Computing actual piece mobility, doing some board scanning to see if that isolated pawn is actually weak, determining if the king is vulnerable if it doesn't castle before assigning that penalty for not castling, and things like that. I don't think a static evaluator will be able to understand more positional ideas as it searches deeper. It just knows that a rook on the 7th rank is better because the programmer added that information into the piece square tables, not because it restricts black in any way. A dynamic evaluator will be able to understand positional ideas better as it searches deeper, and it would move the white rook to the 7th rank because that move would restrict black's pieces, not because the programmer assigned some static bonus for a rook on the 7th rank. Initially, I was thinking in terms of static evaluators, where the "positional knowledge" is all hard coded in static bonuses and penalties, but I could see how a dynamic evaluator could "learn" a positional idea by searching deeper. Am I understanding you right? Also, this brings up another question. Is it better to evaluate things dynamically (by dynamic, I don't mean that a search is done)? It seems like it would be better to evaluate things like weak pawns, king safety, piece activity, and so on, in a dynamic way, because a lot of times the "rule" doesn't apply. IE doubled pawns can be an asset in the early part of a game, castling is usually a good idea, but is not always necessary, and so on.
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.