Author: Dieter Buerssner
Date: 01:28:20 01/02/01
Go up one level in this thread
On January 01, 2001 at 23:02:44, Jon Dart wrote:
[Please refer to original post for board displays]
> 8/8/r1B2k2/4p3/4K2p/5N2/8/8 w - -
>
>Arasan played Nxe5, which Crafty seems to think is losing. It goes to a
>large - score right away:
I tried the 2 position with Yace, with and without TBs (essentially all 4-men +
KPPKP) With TBs I get a draw score very fast in the first position. Without,
the scores stays at about -0.4 after Nxe5. The score just means, that
Yace finds R vs. minor piece, which it considers more or less drawing.
> 2n5/8/1p6/pK5p/2N2k1P/8/PP6/8 w - h6
>
>Here Chester played a4, but the sac Nxb6 seems to be winning.
With TBs, Yace wants to move Kc6 here.
24305346 3:00.4 0.71 15t 1.Kc6 a4 2.a3 Kg3 3.Nd2 Kxh4 4.Kc7 Ne7 5.Kxb6
Kg5 {HT} {0}
As it was allready commented, Nxb6 is drawing.
32006 1.187 0.00 9t 1...Nxb6 2.Kxb6 Kg4 3.a4 Kxh4 4.Kxa5 {EGTB}
4...Kg4 5.b4! h4 6.b5! h3 7.b6! h2 8.b7! {Not
all positions found in TB} {-80}
Without TBs Yace switches to Nxb6 after 17 seconds.
3339997 16.085 0.98 14t 1.a4 Kg4 2.Nxb6 Nxb6 3.Kxb6 Kxh4 4.Kxa5 Kg5
5.b4 h4 6.b5 h3 7.b6 h2 8.b7 h1=Q 9.b8=Q Qc6
{80}
3536976 16.971 0.99 14t+ 1.Nxb6 Nxb6 2.Kxb6 Kg4 3.Kxa5 Kxh4 4.b4 Kg3
5.b5 h4 6.b6 h3 7.b7 Kg2 8.b8=Q h2
3807321 18.080 1.06 14t 1.Nxb6 Nxb6 2.Kxb6 a4 3.Ka5 Kg4 4.Kxa4 Kxh4
5.b4 Kg5 6.b5 h4 7.b6 h3 8.b7 h2 9.a3 h1=Q
10.b8=Q {80}
I recognized something totally unrelated with the last position. In
the FEN the ep. square is given as h6. Obviously blacks last move was
h7-h5. This seems to be in accordance with the formal description of the FEN
by Steven J. Edwards. I remember a similar example from that text, where
the ep. target is set, but an ep. move actually is not possible.
This does not make too much sense to me, but no real problem either.
But it might be the source of two interesting bugs (both of it I had). When
a program reads in the position and really sets the ep square in the internal
representation the repetition logic may fail. When you come to the same
position the third time, the program will fail to see this, because in the
first occurance the ep. square was set, and this, due to chess rules, is
counted as a different position. (Programs, that do not care about
castling rights and ep square for repetition detection may have an advantage
here ...)
The second thing is EGTB probing. Some time ago, my attention was brought to
a position, that was in TBs allready, where Yace totally blunderd. The reason
was, that the ep square was set in a position after a double pawn push, where
the pawn could not be captured ep. In the TB probing code, I also set the
ep square for PfIndexCalc. The TBs returned a mate score, for
a position that is drawn ... After setting the ep square only in cases,
where the pawn really can be captured e.p., this problem has gone.
Regards,
Dieter
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.