Author: Robert Hyatt
Date: 08:30:29 08/28/03
Go up one level in this thread
On August 28, 2003 at 10:18:56, scott farrell wrote: >When you read most literature it says "to find quiet positions" so the eval >function is more accurate, and the quiescent in the name suggests the same. But >I think this is wrong, and most people use it for the later. > >Here are things I and other have put in/out/in/out of their qSearches: >- capture by LVA and SEE >- checks >- passed pawns > >And as we but more into the qSearch, in an effort to stop spending too much >nodes on it, we try: >- pruning >- stand pat - ala Crafty >- etc > >The whole idea of getting to the next ply is to reveals horizon errors and such >in the current ply. > >I have a few positions where a black pawn cant be stopped because of wrong >colour bishop, and a white pawn similarly close and supported and attacked by >bothKings and will never score a try. For the main search it just keeps >horizoning the fact that one pawn is unstoppable. (maybe I need to knowledge for >this instead of search). As soon as I put passed pawn in qSearch, it sees this >in about 6 plies, whereas before it needs something like 15 plies (to exhaust >its best efforts at horizoning it). Checks in the qsearch fall into the same >pile, whereas captures are different. Extensions dont seem to help the >horizoning problem, only the qSearch. > >So it is the qSearch's job to find quiet or to stop some classes of horizon >problems? Those are the same thing. Horizon problems are caused by the evaluation function returning a wrong value, because the position is not "quiet". > >A few threads down, you'll see Robert's note about QxP on the last node, and the >next iteration obviously sees the folly with PxQ. Good captures in the asearch >grab this easily, and I think was the original intent of the qsearch as the >oustanding PXQ is HOT and not quiet. Correct. But by the same token, a position with a pawn on the 7th is also not very "quiet" unless you verify that the pawn can't promote without being lost outright. Either your evaluation has to determine that, or you add stuff to the q-search to determine that, or you accept the occasional error it will cause. > >opinions? > >Scott
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.