Shredder 13 problem

By Emil Vlasák Date 2016-11-09 10:23

Before buying the expensive DeepShredder 13 engine I have visited Jiri Dufek to run several tests.
I have found the following problem.

Lubomir Koblizek, Special hm, Die Schwalbe 2015
1.f7 Rg8 2.fxg8B! Kg1 3.Bg2!.

Every good engine include Shredder 12 is able to find the solution almost immediately.
DeepShredder 13 does not.
Generally it is able to see a bishop underpromotion only on the root.

First it completely disqualifies this engine for using in endgame studies.
Second something like we have already seen, did not we

By Peter Martan Date 2016-11-09 11:24 Edited 2016-11-09 11:40

Pity, you're right, Emil. Shredder "solves" with underpromotion to Bishop after giving in 1.f7 Rg8 with 2.fxg8B correctly, just one move before at the very starting position it doesn't count that in, giving 0.00 for Queen- promotion in output at starting- position.
Yet some more kind of bug but a simple matter of too much pruning, because after Backward of the first move, mate- score isn't kept in hash but 0.00 and Queen- promotion arise again.
Rybka- Fritz15 and the Ippos like Fire 4 still don't know Bishop- underpromotion at all, Houdini2 already did know it and all the other newer engines do know it too.
As for "knowledge", S13 seems to have it coded too, otherwise it couldn't solve such at root- position neither, I guess.
So maybe a special bug as for this certain position?
Will look for other examples.

By Günter Müller Date 2016-11-09 11:46

Hallo Team

10700: Lubomir Koblizek, Special hm,,

1.f7 Rg8 2.fxg8B! Kg1 3.Bg2!.

We have the same with another engines:

8/8/4pP2/4P3/8/7B/3K3p/6rk w - - 0 1

Analysis by Gull 3 x64:

1.f7 Rg8 2.fxg8Q
= (0.00) Depth: 30/50   00:00:00

Analysis by Deep Rybka 4.1 SSE42 x64:

1.f7 Rg8[] 2.fxg8Q
= (0.00) Depth: 28   00:01:25 15498kN, tb=42119

Analysis by Fire 4 x64:

1.f7 Rg8 2.fxg8Q
= (0.00) Depth: 30   00:00:29 67392kN

(, 09.11.2016)

MfG
Günter

By Peter Martan Date 2016-11-09 12:37 Edited 2016-11-09 12:46

Peter Martan schrieb:

Will look for other examples.

As for some more positions with Bishop- underpromotion to be be found at root (first move right after starting- position) Shredder13 has no problem to find, indeed it seems to be a bug (or a feature

) as for follow- up moves only.
Here is another study with two Bishop- underpromotions to be found, giving in the moves up to the first one, S13 finds at root, but the eval stays wrongly a winning one as long as the second Bishop- underpromotion has to be reached yet still as next move to be found too.

Event:

Ort:

Datum:

Weiss:

Schwarz:

Ergebnis

Board

P.S. As known click "Zitieren" to see the lines in braces too.