Author: Vincent Diepeveen
Date: 11:43:30 10/11/02
Go up one level in this thread
On October 11, 2002 at 13:52:04, Uri Blass wrote: Not really. I just do singular extensions and checks in qsearch when we talk about this position. It would amaze me if deep blue didn't use singular extensions, because even chiptest back in the 80s or something already used it :) >On October 11, 2002 at 13:14:55, Vincent Diepeveen wrote: > >>On October 11, 2002 at 11:03:19, Uri Blass wrote: >> >>DIEP needs 9 ply to see the KRP KRP endgame (so winning >>the pawn on e5). > >You must have a lot of extensions to see it only at 9 plies. > >1...h3 2.Rh6 a3 3.Rh8+ Ke7 4.Rh7+ Ke8 5.Rxh3 Ra4 6.Rh8+ Ke7 7.Rh7+ Ke8 >8.Rh1 a2 9.Ra1 Ke7 10.Ke5 Ra5+ 11.Kd4 > >If I count only no check moves I get 14 plies even with checks in the qsearch. > >1...h3(1 ply) >2.Rh6(second ply) >2...a3(third ply) >3...Ke7(4th ply) >4...Ke8(5th ply) >5.Rxh3(6th ply) >5...Ra4(7th ply) >6...Ke7(8th ply) >7...Ke8(9th ply) >8.Rh1(10th ply) >8...a2(11th ply) >9.Ra1(12th ply) >9...Ke7(13th ply) >10.Ke5(14th ply) >rest qsearch. let me count for you more clear instead of the above line, of course also extending check escapes: h3 1 rxh6 2 a3 3 rxh3 4 ra4 5 rh8+ 6 ke7 7 rh7+ 7 kf8 8 rh1 8 a2 9 ra1 10 ke7 11 NULLMOVE 12 qsearch: ra5+ something kxe5 You see it is trivial that it is 12 ply here without extensions if you do checks in qsearch. If you do not do them, then it is 13 + R = 15 for R=2 So it is 13 ply for a stupid fullwidth program and with hashtables 13 ply is very little, even fullwidth here. But more important is that fullwidth with singular extensions like 9 ply. We still didn't hit the subject extending passed pawn moves... For not doing checks in qsearch, you only have to add 1 ply when searching fullwidth. Where you lose 3 ply to nullmove here, you lose just 1 ply to not doing the checks in qsearch if you don't nullmove. That's the trivial observation. So of course i understand why it didn't find it: it was basically hardware and not using hashtables, so it's impossible to search deep, even what bob calls a 'simple piece square table program'. That's how he classified deep thought. I would classify deep blue I and deep blue II as piece square table programs with a few gnuchess enhancements... > >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.