On 12 Jun 2006 16:27:18 -0700, wrote:


Tony M wrote:
On Mon, 12 Jun 2006 16:40:23 -0400, Amarande
wrote:

wrote:
Amarande wrote:
In the following position -

7B/5K1k/8/8/5NN1/8/8/8 w - - 0 1

It is clear to the human mind that there is an easy mate in 5 he

1 Nf6+! (mandatory, as the B cannot be withdrawn) Kh6 (Kxh8 2 Ng6#) 2
Ne4 Kh7 3 Ba1 (or any B move except Bg7) Kh6 4 Bg7+ Kh7 5 Nf6# (or 5 Ng5#).

Junior 9, however, seems to be acting very strangely when given this
position in analysis mode at each step. (Computer settings: AMD 3500+,
Junior 9 assigned 512MB of 1536MB total memory)

* Instead of 1 Nf6+!, Junior accedes to a stalemate draw by evaluating 1
Ba1? as the best move, eval = (0.00). It stays with this rapidly to
depth 63 (where analysis appears to stop) after 00:00:01.

* Instead of 1 ... Kh6, which delays the mate by an additional 3 moves,
Junior considers Black's best move to be 1 ... Kxh8, allowing Ng6. Still
more incomprehensibly, it 1) evaluates 1 ... Kxh8 as "= (0.00)" despite
there being a White mating move immediately, and 2) it thought of Kh6 at
the very beginning (with an eval of +- (7.71) and then discards it in
favor of capturing the B ...

* In the 1 ... Kxh8 line, why does Junior not see 2 Ng6# at all? It
considers, from the very start of analysis all the way to depth 63, the
best White move to be 2 Ng2?, with a clear draw.

* After 1 ... Kh6, Junior correctly sees the mate in four, but a strange
artifact appears:

2.Ne4 Kh7 3.Bd4 Kh6 4.Bg7+ Kh7 5.Nf6# Kh8
+- (#4) Depth: 9 00:00:00 5kN

Wow, Black gets to make a move after he's checkmated? And a move into
check at that?

This same anomaly is repeated in the analysis at White's third move, but
not at his fourth ...

I'm guessing that you have a corrupted KNNBK endgame tablebase file.
That's the only thing I can think of that could cause such weirdness.

No tablebases are installed, so it can't be that.

Any other ideas?

It looks like a bug to me. It appears that Junior always evaluates a
king and two knights as a draw, even if there is a checkmate
available.

Tony

Indeed, Junior (and a few other engines) has this problem in that it
will immediately stop searching if it calculates a "draw by material".
This explains the case with 1.Nf6+ Kxh8.

But it doesn't explain the case of why it thinks 1.Ba1 is the best
move.

jm

Hmm, yes, that is a weird one. The only theory that I can come up
with is that because of the draw by material glitch, all moves in this
position are considered a draw by Junior. All bishop moves and Ng6
are stalemate, and all other moves allow the bishop to be captured,
which is a draw in Junior's eyes. Maybe Junior generates bishop moves
before knight and king moves, and because all moves are considered a
draw, it just happens to pick the first bishop move generated. That's
my uneducated guess.

I have absolutely no idea why Junior has a principal variation which
contains moves beyond mate.

Tony

PS. John, do you know if Johan de Koning is still working on The
King? I really like that engine (and the Chessmaster series in
general), and would be thrilled to see a new version.