Author: Robert Hyatt
Date: 20:23:20 06/06/00
Go up one level in this thread
On June 06, 2000 at 17:55:45, Torstein Hall wrote: >When adding knowledge to a chess program, do you just add bonuses to a score so >a piece combination, for instanse two connected passed pawns on rank 6 add a lot >etc. >Is it not possible to have several "bonuses" kicking in at the same time? And >possibly in opposite directions etc. This is actually the way many of my terms work. IE X for a passed pawn, -Y if it is blockaded. X for 2 connected passed pawns, Y more if the king is too far away or there is little material left, +Z if this condition, -Q if this condition. That is not uncommon... Some terms are simply heavily qualified before they kick in. Others kick in in bits and pieces and have offsetting things if the opponent has pieces/pawns in the right places to help offer counter-chances.. IE I don't think most programs today use an excessive amount of +X for this, -Y for that. My king-safety, for example, is a 3rd order calculation that is composed of pawn shelter, piece placement and material remaining. All three fairly complex terms get factored together to come up with a king safety number. Other terms are nearly as complex, and very few are outright simple things any more. IE I can't imagine a program just saying any longer "A rook on the 7th is +X". Yes, I have seen a few older programs that did this. And yes they have looked foolish when they play Rh7 and the opponent has no pawns on the 7th, and the king is not on the back rank. I think programs of today are beyond that sort of simple evaluation... hopefully... and getting more beyond it all the time... > >I feel that working that way in the endgame you must end up in a lot of >positions where your eval will be just meeningless. In a lot of cases the >endgame hangs on just one tempo, and a static eval adding bonuses etc. will do >you no good! True. But it is non-trivial to simply get the king to "the right place". So there is plenty of room for work. I hope I do reasonably well in endings, but the code is so complex that it is very likely that bugs creep in and cause weaknesses that ought not be there. > >Would it not be better to just have a fast search doing most of the work, at >least in the late endgames? Search can do a lot. But there are some places where search simply can't do it all. It takes balance between search and smarts to make a program play like they do today. > >Or perhaps I just do not understand how the programs work?! ( Not very >unlikely.....) > >Torstein Probably most of us don't. :)
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.