Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: More WAC busts found by Yace. Are these well known?

Author: Jeremiah Penery

Date: 23:53:17 02/12/01

Go up one level in this thread


On February 13, 2001 at 02:02:37, Dann Corbit wrote:

>I have been fooling around with a special compile of Yace, and it got 294/300 on
>WAC at 5 seconds per position on my machine.  I let Yace have the 6 it missed at
>60 seconds per position.  Upon scrutiy, I have found that there are some
>positions in my WAC test set that are clearly in error.
>
>white ( 1): input: st 60
>white ( 1): input: test wac60.ci
>title position 041
>Stored 0 learned positions into hash table
>solution  Ka5 Kc5
...
>     44923   2.386  Mat21  8t  1.b7 Rg5+ 2.Kc6 Rg6+ 3.Kd7 Kxb7 {EGTB} {80}
...
>   2823031  58.990  Mat12 14f. 1.Ka5 Rg1 2.Rf8+ Kb7 3.g8=Q Ra1+ 4.Kb4 Rb1+
>                               5.Ka3 Ra1+ 6.Kb2 Ra5 7.Qg2+ Ka6 8.Qc6 Rb5+
>                               9.Kc2 Rc5+ 10.Qxc5 {EGTB} 10...Kb7 11.Ra8 Kxa8
>                               12.Qc8#! {1081}

>Move b7 is a mate in 21, and yet is not listed among the solutions.  That is
>clearly an error.

I think you posted this before, and it was agreed to be a cook.  However, it was
also said that this test was really written for humans, and that Ka5/Kc5 are the
much better human moves.  Also, they do lead to a mate a lot faster, so they
could be considered better moves than b7.

>title position 100
>Stored 0 learned positions into hash table
>solution  b6+
...
>  35515876  52.292   2.92 20t  1.Kb3 Bd6 2.Kc2 Bc7 3.Kd3 Bd6 4.Ke3 Bc7 5.Bc3
>                               Bd8 6.Kd3 Kb8 7.Kc2 Ka7 8.Kb2 Be7 9.Kb3 Bd8
>                               10.Ka4H Be7H 11.Bd2H {HT}
>  36207255  53.290   2.93 20t+ 1.Ka3 Kb8
>  39780878  58.990   2.93 20t  1.Ka3 Kb8
>  39780878  58.993   2.93 20u. 1.Ka3 Kb8

>Curious about this one.  Are Be3, Ke3, and Ka3 bad moves?  They seem to deliver
>a decisive advantage, but I will admit I have not studied the board position
>recently.  Let's see, here it is:
<fixed diagram, hopefully>
[D]8/k1b5/P4p2/1Pp2p1p/K1P2P1P/8/3B4/8 w - -

This is almost surely computer mis-evaluation.  b6 is really the only way to
break open the position and win, but I think there might be a way to win with
Be3 and playing b6 or something later.  I think someone else posted that an
IM/GM found that solution, but I might remember wrong.

>title position 141
>Stored 0 learned positions into hash table
>solution  Qxf4
...
>This one (WAC 141) was just a little slow.  Took 14 seconds to find the right
>answer and 25 seconds to recognize a checkmate.

This is a big null-move-killer problem.  If you turn null-move off, you should
find it 2-3 plies earlier.

>title position 146
>Stored 0 learned positions into hash table
>solution  Bc8
...
>   2228454  58.997  Mat51 14u. 1.Bd3 Ke3 2.c5 Kxd3 3.cxd6 {EGTB} 3...f2! 4.d7!
>                               f1=Q! 5.d8=Q! Qc1+! 6.Kd7! Qc4 7.d6! Qa4+!
>                               8.Ke7! Qe4+! 9.Kf7 Qd5+ 10.Ke8! Qh5+! 11.Kd7!
>                               Qb5+! 12.Kc8! Qa6+! 13.Kb8! Qb5+! 14.Ka7! Qa4+!
>                               15.Kb7! Qb5+! 16.Qb6! Qd7+! 17.Kb8! Kd2 18.Qc7!
>                               Qe8+! 19.Qc8! Qe5 20.Qc6! Ke2! 21.Qc4+ Kd2
>                               22.Qb4+! Ke2 23.Kb7! Qg7+! 24.Kc6! Qg2+ 25.Kc7!
>                               Qg3! 26.Qc4+! Kf2! 27.Qd5! Qc3+ 28.Qc6! Qg3!
>                               29.Kb7! Ke1! 30.d7 Qb3+! 31.Kc8! Qh3! 32.Qf6!
>                               Ke2! 33.Kc7! Qh7! 34.Qe6+ Kf1! 35.Qd5 Qh2+!
>                               36.Kc6! Qc2+! 37.Kb6! Qf2+! 38.Ka5! Qe1+!
>                               39.Kb5! Qb1+! 40.Kc5! Kg1 41.d8=Q! Qc1+ 42.Qc4
>                               Qa3+! 43.Qb4 Qa7+ 44.Qbb6 Qxb6+! 45.Qxb6! Kg2
>                               46.Qf6 Kh2 47.Qg6 Kh3 48.Kd5 Kh4 49.Ke4! Kh3
>                               50.Kf3! Kh4 51.Qh6# {1001}

>Since Bd3 is a checkmate, it is surely a correct solution to the problem.

This was also discussed a couple times in the past here.  Bc8 leads to mate in
20, and is a much nicer solution.  I guess for a computer that all the solutions
should be included, but for a human Bc8 should be the solution.

>title position 230
>Stored 0 learned positions into hash table
>solution  Rb4
...
>  33737137  58.992   1.87 16f. 1...Rh7 2.Rb1 a4 3.Ba3 Rh5 4.f4 Bb7 5.Bb2 Rh8
>                               6.Ra1 Kb5 7.Kg2 Bc6 8.Rh1 Rb8 9.Kf3 {80}

>Here, Yace really just missed it.  But I think 230 is probably the hardest one
>in the set.

Definitely.  I'm pretty sure Rb4 is the only way to win, but I think it's partly
a null-move thing and also a passed-pawn thing.  I had some version of Crafty
where I increased certain pawn-scoring and extensions a lot and WAC230 was
solved very quickly.

Jeremiah



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.