Author: Michel Langeveld
Date: 14:54:10 03/15/04
Go up one level in this thread
On March 15, 2004 at 04:39:38, Renze Steenhuisen wrote: > >Hi all, > >I wondered, why does my Quiescence Search cause the overall move-ordering to >drop? >I use PVS and a Transposition Table for early cutoff and move suggestion from >the TT. My evaluation function is very simple and I only use static >move-ordering for the moment. > >Any ideas? > >Cheers! > Renze I tested 2 positions: [D]rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - With QSearch PV:1.e4 e5 2.Nf3 Nf6 3.Nc3 Nc6 4.Bb5 Bd6 5.o-o o-o score:0.05 d:10 sec:10.80 nod es:2248k nps:208k FH:83.1% qnodes:806k qperc:35.9% evals:1664k 0cuts:297k moves: 2392k chkext:18k hprobes:3148k hhits:220k hhit%:7.0% hstores:841k hfull:15.3% pp robes:1664k phits:1385k phit%:83.2% phstores:279k phfull:39.2% Without QSearch PV:1.d4 Nf6 2.Nf3 e6 3.h3 Nc6 score:-0.08 d:10 sec:4.78 nodes:1298k nps:272k FH: 90.2% qnodes:0 qperc:0.0% evals:752k 0cuts:397k moves:1349k chkext:11k hprobes:2 050k hhits:52k hhit%:2.6% hstores:115k hfull:2.3% pprobes:752k phits:635k phit%: 84.4% phstores:116k phfull:19.3% FH is indeed worse with QSearch [D]4r1k1/p5p1/2p4p/4rp1P/P4Q2/2RbBPP1/q7/4R1K1 b - - With QSearch PV:1... Qd5 2.Rcc1 Bc2 3.Rxc2 Rxe3 4.Rf1 Kh8 5.Rb2 Re1 6.Rb1 score:1.21 d:10 sec :31.81 nodes:10877k nps:342k FH:92.7% qnodes:3611k qperc:33.2% evals:8486k 0cuts :827k moves:13287k chkext:273k hprobes:15963k hhits:994k hhit%:6.2% hstores:2876 k hfull:45.6% pprobes:8486k phits:6765k phit%:79.7% phstores:1721k phfull:19.7% Without QSearch PV:1... Qd5 2.Rcc1 Qb3 3.Qd4 Bc2 4.Kh2 Qxa4 5.Qxa4 score:0.78 d:10 sec:29.27 nod es:13491k nps:461k FH:94.8% qnodes:0 qperc:0.0% evals:8617k 0cuts:3044k moves:16 589k chkext:411k hprobes:22104k hhits:816k hhit%:3.7% hstores:1217k hfull:20.2% pprobes:8617k phits:7752k phit%:90.0% phstores:864k phfull:13.2% FH is better without QSearch! I was thinking about this... it can be if you order your moves based on mvvlva (Most Valuable Victim Least Valuable Attacker) that this order is more right without QSearch then with QSearch. If a qsearch is done ... then the value is more close to a SEE value. So maybe if you add SEE the opposite is true.
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.