Author: Bruce Moreland
Date: 10:11:24 11/27/00
Go up one level in this thread
On November 27, 2000 at 05:26:14, Gian-Carlo Pascutto wrote: >On November 27, 2000 at 03:26:16, Bruce Moreland wrote: > >>It always blows up if you don't try to stop it from blowing up. >>If people survive beyond this initial realization, they realize that you have >to do something about qsearch or you get stuck in ply 4 or so. >> >>I had these problems five years ago. The version of my program that finished >>3rd in the 1995 WMCCC in Paderborn did something like 80% quiescent nodes. > >This sounds so familiar, except that I had it with crazyhouse and >check extensions. > >The first version of my program simply did 1-ply-on-check. It got >killed often because the search totally blew up and it had to make >moves that were the result of a nominal 2 ply search. > >Problem is, you can't just disable the extension. But I can't >lower it because I don't have fractional extensions (yet), and >because the extension IS really important in crazyhouse. > >So what do you do? You try to hack in some more intelligence in >the extension. You try to avoid doing it if it gets big unless >it's absolutely necessary to stablize the search. It helps. A bit. >It now makes moves that are based on nominal 3-ply searches... Add fractional extensions, and extend 3/4 ply on single response to check. Also, don't do it when the side that is materially winning (big) is the one doing the checking. bruce >But it spots mate in 6 during those searches. That makes >it reasonable. > >My quiescent search is more or less the same story. It often does >more than 80 or 90% qnodes. But it's reasonable most of the time. > >-- >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.