Author: Robert Hyatt
Date: 06:48:23 01/08/04
Go up one level in this thread
On January 08, 2004 at 07:31:10, Mike Hood wrote: >On January 07, 2004 at 10:44:36, Robert Hyatt wrote: > >>On January 07, 2004 at 05:26:58, Mike Hood wrote: >> >>>On January 07, 2004 at 00:20:17, K. Burcham wrote: >>> >>>> >>>>Steve Maughan, big book of combinations >>>> >>>>Tibensky - Franzen >>>>Find Kh6 >>>> [D] 1r3rk1/b1pR1p2/p3p1p1/2q1P1Kp/P1N2P2/1P1Q2P1/7P/3R4 w - - >>> >>>This is on my humble Pentium III-733. Patzer was the first engine I tried, >>>because it seems to be skilful at finding the right move in closed positions. >>> >>>Analysis by Patzer 3.11: >>> >>>1.Qxg6+ >>> -+ (-10.00) Depth: 1 00:00:00 >>>1.Qf3 >>> ² (0.42) Depth: 1 00:00:00 >>>1.Qf3 Bb6 >>> µ (-0.98) Depth: 2 00:00:00 >>>1.Qf3 >>> µ (-0.96) Depth: 2 00:00:00 >>>1.Kh4 >>> ³ (-0.57) Depth: 2 00:00:00 >>>1.Kh4 >>> = (-0.18) Depth: 3 00:00:00 >>>1.Qf3 >>> = (-0.06) Depth: 3 00:00:00 >>>1.Qf3 a5 >>> µ (-0.75) Depth: 4 00:00:00 1kN >>>1.g4 hxg4 2.Kxg4 a5 3.Qh3 Rbe8 >>> = (-0.04) Depth: 4 00:00:00 9kN >>>1.g4 hxg4 2.Kxg4 Ra8 3.Qg3 >>> = (0.17) Depth: 5 00:00:00 36kN >>>1.g4 hxg4 2.Kxg4 a5 3.Qh3 Rfe8 4.Rxf7 Kxf7 5.Qh7+ >>> = (-0.06) Depth: 6 00:00:01 115kN >>>1.Qf3 Bb6 2.g4 hxg4 3.Qg3 a5 >>> = (-0.03) Depth: 6 00:00:02 152kN >>>1.Kh4 Qf2 2.Rd2 Qc5 3.Rd1 >>> = (0.00) Depth: 6 00:00:02 167kN >>>1.Kh6 >>> ² (0.51) Depth: 6 00:00:03 284kN >>>1.Kh6 a5 2.Nxa5 Qxa5 3.Qxg6+ fxg6 4.Rg7+ >>> ² (0.56) Depth: 7 00:00:04 376kN >>>1.Kh6 a5 2.Nxa5 Qxa5 3.Qxg6+ fxg6 4.Rg7+ Kh8 5.Rh7+ Kg8 6.Rg7+ >>> ² (0.56) Depth: 8 00:00:05 497kN >>>1.Kh6 a5 2.Nxa5 Qxa5 3.Qxg6+ fxg6 4.Rg7+ Kh8 5.Rh7+ >>> ² (0.56) Depth: 9 00:00:08 841kN >>>1.Kh6 Rfc8 2.Rxf7 Kxf7 3.Qxg6+ Ke7 4.Qf6+ Ke8 5.Nd6+ cxd6 6.Rxd6 Rb7 7.Qh8+ Ke7 >>>8.Qf6+ >>> ² (0.56) Depth: 10 00:01:15 11987kN >> >> >>It isn't seeing what it is supposed to see. Kh6 is a big tactical win. Here >>for some reason, patzer simply can't see any way for white to improve after Kh4, >>which is strange as white is better. But since it drops to draw for Kh4, Kh6 >>is chosen for the wrong reason... Most programs evaluate Kh4 and Kh6 as nearly >>the same until they notice that Kh6 wins big material after a pretty deep >>search. > >Maybe you're right, Robert, that Patzer is choosing the right move for the wrong >reason. When setting up Patzer to analyse several lines simultaneously I saw >that it rated several other moves as almost as good as Kh6 (Nd2, Na5, h3, g4, >b4). > >Amusingly, when Patzer gets to search depth 11 it starts accessing the >tablebases. I've included the next two lines of analysis below. I have all the >5-piece EGTBs, so in the 22 moves Patzer must have found a line with 17 captures >:) > >1.Kh6 > ± (0.90) Depth: 11 00:03:03 28901kN, tb=4 >1.Kh6 Qc6 2.Na5 Qa8 3.Qxa6 Qf3 4.Qd3 Qg4 5.Nc6 h4 > +- (1.54) Depth: 12 00:22:48 195857kN, tb=17 That is the point where it is beginning to see what is going on. It is not uncommon for a program to find the right move for the wrong reason. Mine does it all the time, in fact...
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.