Author: Vincent Diepeveen
Date: 14:40:17 04/15/03
Go up one level in this thread
On April 15, 2003 at 17:23:09, Will Singleton wrote: Will, Some programs have a very simple qsearch and simply recapture at all squares where already in the normal search has been captured. Simple, easy and very fast. Note that scores are from blacks viewpoint seen in diep: 00:00 95151 95k 0 0 51382 (2) 5 (25,69) -0.941 Nc6-e5 Qb3-b4 Nf6-g4 g2-g3 Ne5- d3 Bc4xd3 e4xd3 00:00 110310 110k 0 0 88248 (2) 6 (27,97) -0.941 Nc6-e5 Qb3-b4 Nf6-g4 g2-g3 Ne5- d3 Bc4xd3 e4xd3 00:02 111874 112k 0 0 247243 (2) 7 (41,163) -0.635 Nc6-e5 Qb3-b4 a7-a5 Bc5-b6 a5 xb4 Bb6xc7 Rd8xd4 e3xd4 Ne5xc4 c3xb4 00:04 124403 124k 0 0 566037 (2) 8 (47,283) -0.635 Nc6-e5 Qb3-b4 a7-a5 Bc5-b6 a5 xb4 Bb6xc7 Rd8xd4 e3xd4 Ne5xc4 c3xb4 ++ c6-a5 procnr=1 terug=-634 org=[-635;-634] newwindow=[-635;520000] 00:08 136815 137k 0 0 1223129 (2) 8 (113,503) 0.630 Nc6-a5 Bc4xf7 Qc7xf7 Qb3-b4 Qf7-c4 Bc5-e7 Rd8-d5 Rf1-c1 00:11 146045 146k 0 0 1718953 (2) 9 (141,719) 0.822 Nc6-a5 Bc4xf7 Qc7xf7 Qb3-b4 Na5-c4 Nd4-c6 Rd8-d2 Bc5xa7 Nf6-g4 00:23 151133 151k 0 0 3486654 (2) 10 (206,1456) 0.105 Nc6-a5 Qb3-b4 Na5xc4 Nd4-b 5 Qc7-e5 Qb4xc4 Nf6-g4 Rf1-d1 Bc8-e6 Rd1xd8 Ra8xd8 Qc4-b4 Qe5xh2 Kg1-f1 00:45 157337 157k 0 0 7179296 (2) 11 (295,2173) 0.064 Nc6-a5 Qb3-b4 Na5xc4 Nd4-b 5 Qc7-e5 Qb4xc4 Nf6-g4 g2-g3 Bc8-e6 Qc4-b4 a7-a5 Qb4-a3 ++ c6-e5 procnr=1 terug=65 org=[64;65] newwindow=[64;520000] 02:32 156889 157k 0 0 23997827 (2) 11 (337,2598) 0.383 Nc6-e5 Qb3-b4 Ne5xc4 Nd4- b5 Qc7-e5 Qb4xc4 Nf6-g4 g2-g3 Bc8-e6 Qc4-b4 a7-a5 Qb4-a3 Rd8-d2 >[D]r1br2k1/p1q2ppp/2n2n2/2B5/P1BNp3/1QP1P3/5PPP/R4RK1 b - - > >Some programs never show an elevated score, others show +2 or so before figuring >it out. It has to do with a piece that is guarding a pinned piece at the end of >the pv. Ruffian and Yace 99.5 exhibit the problem, while Tao doesn't. I'm >wondering if this problem is due to a lack of extensions in the pv, or a qsearch >artifact, or something else. > >ruffian: >9 119 220 3044077 Ne5 Qb4 Nxc4 Qxc4 Rd5 Nb3 Rb8 Rad1 Rxd1 Rxd1 Rxb3 Qxb3 Qxc5 >10 224 695 9490320 Ne5 Qb4 Nxc4 Nb5 Qc6 Nxa7 Qa6 Nxc8 Raxc8 Rfd1 Nd5 >11 36 1278 16931060 Ne5 Qb4 Nxc4 Nb5 Qc6 Qxc4 Nd7 Rad1 Bb7 Rxd7 Rxd7 Nxa7 > >If you look at Ruffian's ply 10 pv with a +2.24 score (Ne5 Qb4 Nxc4 Nb5 Qc6 Nxa7 >Qa6 Nxc8 Raxc8 Rfd1 Nd5), white plays Nxa7, giving up the bishop for a pawn. >The reason can be seen in Amateur's ply 10 pv below, which is that if Qxc4 >instead of Nxa7, the bishop looks lost due to Rd5, without any compensation. > >10 216 736 3122242 Ne5 Qb4 Nxc4 Nb5 Qc6 Qxc4 Rd5 Rd1 Rxc5 Rd8 Ne8 > >However, at the end of that pv, you get the following position: > >[D]r1bRn1k1/p4ppp/2q5/1Nr5/P1Q1p3/2P1P3/5PPP/R5K1 w - - > >Clearly, white can just take the rook for free. Apparently, a lot of programs >cannot see this. Just speculating: in the qsearch the move Qxc5 would be tossed >out, since a typical SEE would score it a losing capture. If this is indeed the >problem, it would be helpful to know how some programs avoid it. It's possible >that some programs just extend the relevant pv more than others. Mine only >would have 1 extension, since the recapture Qxc4 doesn't follow the first >capture. Or, perhaps there's a qsearch method that doesn't toss out Qxc5. > >Will
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.