Author: Mridul Muralidharan
Date: 06:32:55 06/14/04
Go up one level in this thread
On June 14, 2004 at 02:06:17, Uri Blass wrote: >On June 14, 2004 at 01:25:57, Mridul Muralidharan wrote: > >>Hi all, >> >> I have been following the games of my engine and one very painful thing is to >>see it fall from +4.x to 0.0 due to a king hunt by opponent. >>Some programs like yace , frenzee (atleast in one the games I saw) , etc seem to >>identify this correctly and immediately take the eval down to 0.0 - this is not >>a search depth thing. >>My engine on the other hand needs a few moves to see the inevitability of the >>position. >>I am sure it might not be always possible to detect a king hunt always with 100% >>accuracy , but any general tips , comments to get me started would be great ! >> >> >>Thanks >>Mridul >> >>PS : Here is a game at icc today morning where it happens again from my winboard >>saved game : the comments are also present which give an idea of the evals by >>both engines. >> >>For my initial look : >> >>[D] 6rk/pnp2p1n/1p1p3p/3PpN1q/4P1r1/2P1Q1B1/P1P2PP1/1R2R1K1 b - - 0 24 >> >>Here , Ng5 gets played since borer does not understand that the position might >>lead to a draw ! > >I do not think that the game moves after Ng5 are forced from black point of >view. > >30...Qh1+ lead to a draw but yace suggests 30...Rg3 with evaluation of advantage >for black > >PostModernist - witchess, ICS rated standard match 2004 >[D]7k/pnp2p2/1p1p3p/3Pp3/4P3/2P1QP1q/P1P1R1r1/1R2K3 b - - 0 1 > >Analysis by Yace Paderborn: > >30...Rxe2+ 31.Kxe2 > ± (0.88) Depth: 1 00:00:00 >30...Rh2 31.Rxh2 Qxh2 > ± (0.80) Depth: 1 00:00:00 >30...h5 31.Rxg2 Qxg2 > ± (0.76) Depth: 1 00:00:00 >30...Qg3+ 31.Kd2 > ² (0.65) Depth: 1 00:00:00 >30...Qg3+ 31.Kd2 > ² (0.65) Depth: 1 00:00:00 >30...Qh1+ 31.Qg1 > ² (0.29) Depth: 1 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 > -+ (-3.15) Depth: 1 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 > -+ (-3.15) Depth: 1 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 > -+ (-3.15) Depth: 2 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 > -+ (-3.15) Depth: 2 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 > -+ (-3.15) Depth: 3 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 > -+ (-3.15) Depth: 3 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 34.Qg5+ Kf8 > -+ (-3.31) Depth: 4 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 34.Qg5+ Kf8 > -+ (-3.31) Depth: 4 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 34.Qg5+ Kf8 35.Qh6+ Ke8 36.Qh8+ >Kd7 > -+ (-3.57) Depth: 5 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 34.Qg5+ Kf8 35.Qh6+ Ke8 36.Qh8+ >Kd7 > -+ (-3.57) Depth: 5 00:00:00 >30...Qh1+ 31.Kd2 Rxe2+ 32.Qxe2 Qxb1 33.Qa6 Nc5 34.Qc8+ Kg7 35.Qxc7 Qxa2 36.Qxd6 > -+ (-3.43) Depth: 6/18 00:00:00 41kN >30...Qh1+ 31.Kd2 Rxe2+ 32.Qxe2 Qxb1 33.Qa6 Nc5 34.Qc8+ Kg7 35.Qxc7 Qxa2 36.Qxd6 > -+ (-3.43) Depth: 6/19 00:00:00 63kN >30...Qh1+ 31.Kd2 Rxe2+ 32.Qxe2 Qxb1 33.Qa6 Nc5 34.Qc8+ Kg7 35.Qxc7 Qxa2 36.Qxd6 > -+ (-3.43) Depth: 7/19 00:00:00 92kN >30...Qh1+ 31.Kd2 Rxe2+ 32.Qxe2 Qxb1 33.Qa6 Nc5 34.Qc8+ Kg7 35.Qxc7 Qxa2 36.Qxd6 > -+ (-3.43) Depth: 7/26 00:00:00 188kN >30...Qh1+ 31.Kd2 Rxe2+ 32.Qxe2 Qxb1 33.Qa6 Qg1 34.Qxb7 Qf2+ 35.Kc1 Qxf3 > -+ (-3.03) Depth: 8/26 00:00:00 237kN >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 34.Qg5+ Kf8 35.Qh6+ Ke8 36.Qh8+ >Kd7 37.Qh3+ Ke7 38.Qh4+ f6 39.Qh7+ Kd8 40.Qg8+ Kd7 41.Qe6+ Kd8 42.Qxf6+ Kd7 >43.Qf7+ Kc8 44.Qe6+ Kd8 > -+ (-2.65) Depth: 8/26 00:00:01 304kN >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 34.Qg5+ Kf8 35.Qh6+ Ke8 36.Qh8+ >Kd7 37.Qh3+ Ke7 38.Qh4+ f6 39.Qh7+ Kd8 40.Qg8+ Kd7 41.Qe6+ Kd8 42.Qxf6+ Kd7 >43.Qf7+ Kc8 44.Qe6+ Kd8 > -+ (-2.65) Depth: 8/30 00:00:01 623kN >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 34.Qg5+ Kf8 35.Qh6+ Ke8 36.Qh8+ >Kd7 37.Qh3+ Kd8 38.Qh4+ Kc8 39.Qg4+ Kb8 40.Qg8+ Nd8 41.Qxd8+ Kb7 > -+ (-2.25) Depth: 9/30 00:00:02 679kN >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 34.Qg5+ Kf8 35.Qh6+ Ke8 36.Qh8+ >Kd7 37.Qh3+ Kd8 38.Qh4+ Kc8 39.Qg4+ Kb8 40.Qg8+ Nd8 41.Qxd8+ Kb7 > µ (-1.25) Depth: 9/31 00:00:02 707kN >30...Qh1+ 31.Kd2 Rxe2+ 32.Kxe2 Qxb1 33.Qxh6+ Kg8 34.Qg5+ Kf8 35.Qh6+ Ke8 36.Qh8+ >Kd7 37.Qh3+ Kd8 38.Qh8+ Ke7 39.Qh4+ Kd7 40.Qh3+ Kd8 41.Qh8+ Ke7 42.Qh4+ Kd7 >43.Qh3+ > = (0.00) Depth: 9/32 00:00:02 761kN >30...Rg3 31.Kd2 Rxf3 32.Qxh6+ Qxh6+ 33.Re3 > = (-0.01) Depth: 9/32 00:00:03 1451kN >30...Rg3 31.Kd2 Na5 32.Qd3 Rxf3 33.Qb5 Rxc3 34.Qe8+ Kg7 35.Rg1+ Rg3 36.Re3 > ³ (-0.53) Depth: 9/32 00:00:05 2271kN >30...Rg3 31.Kd2 Na5 32.Qd3 Rxf3 33.Qb5 Rxc3 34.Qe8+ Kg7 35.Rg1+ Rg3 36.Re3 > ³ (-0.53) Depth: 9/32 00:00:07 3085kN >30...Rg3 31.Kd2 Rxf3 32.Qg1 Qh4 33.Qh2 Qg5+ 34.Ke1 Nc5 35.c4 Qg3+ 36.Qxg3 Rxg3 >37.Rb5 > ³ (-0.52) Depth: 10/32 00:00:11 4393kN >30...Rg3 31.Kd2 Rxf3 32.Qg1 Qh4 33.Qh2 Qg5+ 34.Ke1 Nc5 35.c4 Qg3+ 36.Qxg3 Rxg3 >37.Rb5 > ³ (-0.52) Depth: 10/33 00:00:17 7304kN >30...Rg3 31.Kd2 Na5 32.Qd3 Rxf3 33.Qa6 Rxc3 > ³ (-0.56) Depth: 11/33 00:00:27 11504kN >30...Rg3 31.Kd2 Na5 32.Qd3 Rxf3 33.Qa6 Rxc3 > ³ (-0.56) Depth: 11/39 00:00:45 19193kN >30...Rg3 31.Kd2 Rxf3 32.Qg1 Nc5 33.Rbe1 Rg3 34.Qh1 Qxh1 35.Rxh1 Kg7 36.Rh5 a5 > ³ (-0.63) Depth: 12/39 00:01:10 29481kN > >(Blass, Tel-Aviv 14.06.2004) > >Uri You missed my point . The point is - if my engine understood that this position leads to a draw , it wont play Ng5 in first place. In that entire variation , it could have tried something different : since according to it , black was up in this position. It sees that Ng5 leads to the variation played in the game , and thinks it is a great position to reach - and without a king hunt / draw detection in evaluation , even after Qh1+ , it does not see a draw. I know yace solves it - and I think I already mentioned it : this is a eval issue as far as I am concerned - you have to see in your eval that it is leading to a draw : by search - there is not draw by either repetition or 50 move rule for a long time to come ! Mridul
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.