Author: José Carlos
Date: 06:16:46 11/21/01
Go up one level in this thread
On November 21, 2001 at 08:48:25, Gian-Carlo Pascutto wrote: >On November 21, 2001 at 03:11:32, José Carlos wrote: > >> I used to limit it in the first versions of my program. Some day I commented >>here and Christophe told me something like "there must be something broken in >>your program; don't limit the qsearch, you'll save nodes because of the better >>information". > >I fail to see how a 32 ply quiescent search will save nodes by giving >better information to a 2 ply nominal search. It's a benefit-general-case decision. Of course you can find weird positions where 30 (never 32) captures will follow but, how many of these do you find in a real game? The "better information" thing means, I belive, better move ordering. >> I fixed my problem (move ordering) and have been working perfectly since then. >>I also did extensive testing some time ago and non-limiting-qsearch version got >>better results. >I'll test it in my program too (not now, it's on the dissection table :) >but keep in mind that there are several ways to limit the quiescent search, >and quite a bit of them are wrong. What I used to do is limit depth. And that proved wrong _in_my_program_. Best thing to do is testing. And best testing method is a combination of games+tests. I have a machine running games night and day (I can't afford two to test pondering). This proves really useful. I've noticed a 300 elo improvement since v0.32 and, for the different betas of v0.33, I see a few points increase. Not perfect, but very useful. José C. >-- >GCP
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.