Author: Don Dailey
Date: 14:12:50 04/22/98
Go up one level in this thread
On April 22, 1998 at 16:17:21, Stuart Cracraft wrote: >I've never been satisfied with my program's >handling of king safety. It has these few factors >combined linearly: > > king on open/half-open file > king next to open/half-open files > king centrality in ending > king non-centrality in non-ending > king near any pawn masses in endgame > king near friendly pawn masses in non-endgame > attacks to squares immediately next to king > >The totality of these is generally far less than a >pawn since I use millipawns and mostly small values >for bonuses/penalties (usually 1 millipawn to 200 millipawns). >Perhaps I should let king safety total more or less than >+1/-1 pawn? > >Anyway, I'm looking to see if there are some good >heuristics I've left out. > >--Stuart Hi Stuart, This is something I've always struggled with too. I do something in my chess program you might try, if you haven't already tried. I think a lot of others do the same type of thing but I don't know this for sure. Each king safety term has a small value like 1 2 or 3. I add them all up in linear fashion and use this as an index into a table that is NOT linear. The idea is that having 3 bad king safety "things" together is a lot worse than their individual sums. I combine this also with normal linear scoring so that I can discriminate badness if only 1 or 2 things exist. In my scheme, having several king safety problems might add up to .20 or .30 of a pawn linearly, but having several simultaneously can make the score go over 1 pawn. I avoid good king safety terms and just semantically view good things as bad things (Bad NOT to have a pawn near the king instead of good to have a pawn near the king.) But additionally, your bad king safety tally should include some terms not normally considered a part of king safety, but probably already measured by your program. A couple examples are presence of enemy queen (anywhere) and tally points if the enemy is better developed than you are. The principle here is that your king is seldom in danger if you are better developed or have intiative and trading queens will dissapate 80% of the attack. Of course any thing you come up with will have exceptions so I'm afraid you'll always struggle with king safety as every program does! I don't claim this is a good way to do it! It's just an idea for you and is basically what I've found to be useful in Cilkchess. I hope you find some of these ideas useful. - Don
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.