Author: Eelco de Groot
Date: 18:48:13 01/06/06
Go up one level in this thread
On January 06, 2006 at 12:01:46, Eelco de Groot wrote: >On January 06, 2006 at 10:54:27, Eelco de Groot wrote: > >>In CSS-Forum Christoph Fieberg reports results with Rybka on Swiss Test 2 at a >>long time control: >> >>http://f23.parsimony.net/forum50826/messages/139718.htm >> >>Rybka can solve 60 out of 64! Not bad. The four remaining positions 7, 12, 40 >>and 49 are maybe not clear, I intend to test them a little more. In Position 7 >>the move to avoid Bxc3 seems to lose but other moves also get a drop in score >>over time. It is easier to post a diagram here, I thought I'd post some results >>here first. First number 7, rest will maybe follow later. I did a four best >>moves analysis with Gambit Fruit 1.0 Beta 4bx, Bxc3+ comes in second but Be7 not >>much better: >> >>am Bxc3+; hmvc 0; fmvc 19; c0 "?"; id "SwissTest2_Nr.07 - Aristarch45-List512"; >> >>[D] r1bq1rk1/p4pp1/5n1p/n2pNP2/1bpP3B/2N1PQ2/1PB3PP/2R1K2R b K - 0 1 >> >>01:31:01.0 -1,65 18 1845764238 Be7 Bxf6 Bxf6 Nxd5 Bxe5 dxe5 Bb7 Rd1 Qh4+ g3 Qh3 >>Qe4 Kh8 e6 Rfe8 Kf2 Rab8 f6 g6 e7 Bc6 Qe5 >>01:31:01.0 -1,75 18 2879197170 Bxc3+ bxc3 Nb3 Rd1 Qb6 O-O Rd8 Qg3 Nh5 Qh3 Nf6 >>Rb1 Rb8 Qg3 Nh5 Qf3 Nf6 Kh1 Qd6 Qg3 >>01:31:01.0 -1,79 18 2139520513 Qb6 O-O Bxc3 bxc3 Nb3 Rcd1 Bb7 Rb1 Ba6 Qg3 Bc8 e4 >>Nxd4 cxd4 Qxd4+ Kh1 Nh5 Qg4 Qxe5 Qxh5 dxe4 >>01:31:01.0 -1,89 18 3337609714 Qd6 O-O Bxc3 bxc3 Nb3 Rcd1 Bb7 Nxc4 dxc4 Qxb7 Qd5 >>Qa6 Rfb8 Rf4 Rb6 Qa4 Kh8 e4 Qd6 e5 >> >>Athlon 2009 MHz, 200 MB HT >> > >At nineteen ply, Bxc3+ is in first again! So this position seems unclear for an >an "avoid move" test. Rybka probably chose the right move afer all. On shorter >timecontrols, -the test is intended for a one minute per move test, at least >that is how I used it-, which move gives best chances may be not so clear. But >that does not make this the best possible testmove. > >03:11:01.6 -1,70 19 1070528192 Bxc3+ bxc3 Nb3 Rd1 Qb6 O-O Rd8 Qg3 Nh5 Qh3 Nf6 >Rb1 Rb8 Qg3 Ba6 e4 dxe4 Rfe1 Kf8 Bxe4 Rxd4 cxd4 Qxd4+ Kh1 Nxe4 >03:11:01.6 -1,79 19 1585142163 Be7 Bxf6 Bxf6 Nxd5 Bxe5 dxe5 Bb7 Rd1 Qh4+ g3 Qh3 >Qe4 Kh8 e6 Rfe8 f6 g6 e7 Nb3 Qxc4 Qg2 Rf1 Ba6 >03:11:01.7 -1,87 19 2163812257 Qb6 O-O Bxc3 bxc3 Nb3 Rcd1 Rb8 Rf2 Rd8 Qf4 Re8 e4 >Bb7 exd5 Bxd5 Bxf6 Qxf6 Nd7 Qd8 Nxb8 Qxb8 Qxb8 Rxb8 f6 g6 >03:11:01.7 -1,98 19 2826541678 Qd6 O-O Bxc3 bxc3 Rb8 Bxf6 Qxf6 Qxd5 Nb3 Rce1 Bb7 >Qd7 Rfc8 Qa4 Bc6 Qa3 Rb7 Nxc4 Bd5 Bxb3 Bxc4 Bxc4 Rxc4 The computer was still running, and the order of the four moves has changed again. Although it not so clear anymore which move is best, the most interesting thing to see in this position is maybe that variability in evaluations can still upset the moveorder in deep searches, and this variability in this case does not seem to diminish with greater depth. To investigate how much of this could be "random noise" as in Chrilly Donninger's theory, maybe you could repeat the experiment with other programs and see if the same moves have similar scores at same depths. It can differ per program of course. The advantage of using multivariation analysis is that you can get four or more PVs from essentially a single position so variability in position does not play such a big role. Here lies some subject material for an IGCA paper maybe? 11:14:36.0 -1,69 21 4185487258 Be7 Bxf6 Bxf6 Nxd5 Rb8 Nxf6+ Qxf6 Rb1 Qd6 O-O f6 Ng6 Re8 Rfd1 Nb3 Be4 Rb5 Kh1 Bd7 g4 Kh7 Qg2 Rb4 Nf4 11:14:36.0 -1,85 20 310254083 Kh8 Ng4 Be7 Bxf6 Bxf6 Nxf6 gxf6 Qxd5 Rb8 Rb1 Bb7 Qxd8 Rfxd8 Kf2 Nc6 b3 Nb4 Be4 Ba6 bxc4 Bxc4 Rhc1 Kg7 Ra1 Rd7 11:14:36.0 -1,97 20 1299140175 Qb6 O-O Bxc3 bxc3 Nb3 Rcd1 Rd8 Qg3 Bb7 Rb1 Qd6 Bxb3 cxb3 Bxf6 Qxf6 Rxb3 Bc8 Qg4 Ba6 Ra1 Bc8 e4 dxe4 11:14:36.0 -2,00 20 1326699232 Bxc3+ bxc3 Nb3 Rd1 Qb6 O-O Rd8 Qg3 Nh5 Qh3 Nf6 Ng4 Nxg4 Bxd8 Qxd8 Qxg4 Qg5 Qxg5 hxg5 e4 Bb7 exd5 Bxd5 Rde1 Rb8
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.