Author: Robert Hyatt
Date: 20:29:15 03/09/03
Go up one level in this thread
On March 09, 2003 at 16:03:45, Uri Blass wrote: >On March 09, 2003 at 15:43:37, Russell Reagan wrote: > >>I think that most people subscribe to the school of thought that says that >>quiescent search is not perfect, so do it fast and "good enough" for most >>situations. If you could have a perfect quiescent search, what price would you >>be willing to pay? One ply of full width search? Two ply? Time to depth takes >>twice as long? I am interested what programmers with more experience than myself >>think about this. > >If I have perfect quiescent search that give me exact evaluation of every >position(win,draw or loss) then even paying 10 plies may do you unbeatable. > >I do not know what you mean by perfect qsearch. The definition is trivial. By definition the _only_ task for the q-search is to take a position given to it by the normal search, and reduce it in simplicity until there is _no_ dynamic characteristics left. If that can be done, then a "static evaluator" can give an accurate score that won't be mislead by tactical considerations nor by positional traps. Not possible to write a perfect one of course. :) > >I think that the idea of Bas Hamastra to do checks in the first plies of the >qsearch is productive. > >Bob tried a similiar idea that did not work but I understood that he only tried >checks in all plies of the qsearch. No. I did as I did in Cray Blitz, which has been documented in the past in several papers: If, at a ply P in the q-search, the following conditions are met, I will try to generate and search checking moves: 1. At all q-search nodes, P-1, P-3, ..., P-n, my opponent was in check, 2. plus at a small number of nodes in the basic search he was also in check (this to be sure that I don't follow a bunch of junk checks in the q-search if the main search wasn't looking at tactical stuff too.) If, at any P in the q-search, and 1. at ply P-2, P-4, ..., P-n, this side is in check; 2. The current side was in check in a small number of plies in the normal search. Then I would try all check evasions by using the normal Search() code which would recognize mates. > >I found that checks in the first plies of the qsearch did movei slightly better >in test suites with no significant demage in games and my function to detect >checks in the qsearch is a slow function that I need to improve. > >I think that other ideas may be also productive and detecting cases when the >side to move is in trouble is important(checks are good candidates but it is >possible to catch more cases and movei catch more cases but not enough). > >Uri
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.