Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Shredder selectivity fails - HERE'S A WORKAROUND!?

Author: William Penn

Date: 16:45:50 06/05/05

Go up one level in this thread


On June 04, 2005 at 13:07:29, Ernest Bonnem wrote:

>[D]6n1/3p4/8/1P1Pk1n1/4P3/B5K1/8/8 w - -
>
>White to win
>
>Here Shredder 9 UCI cannot find the "simple"  b6! even at depth 36/61 (searching
>more than 2 hours on my computer)
>
>b6 is found almost immediately by all other programs I tried (Fritz 7, Tiger,
>Pharaon,...)
>
>Has anybody an explanation ?

D 6n1/3p4/8/1P1Pk1n1/4P3/B5K1/8/8 w - -

I played around with this position using various engines and confirmed that most
engines don't have a problem finding 1.b6 unless you go back as far as Fritz
5.32.

I also tried Shredder 7.04, Shredder 8, and Shredder 9 UCI in both the CB and
Shredder GUIs. They do indeed have a problem finding 1.b6 if you are only doing
one line of analysis. However they find it quickly if you increase the number of
lines being analyzed!

So with Shredder 9 UCI in the Shredder GUI, press "f6" once or twice to expand
to 2-3 lines of analysis in infinite mode and 1.b6 quickly bubbles to the top of
the list. You can then press "f5" to reduce to one line of analysis again, and
it still sees 1.b6 as best. Of course by then it has probably been recorded in
the shredder.pl2 file, so to run another valid test you'll have to start with a
fresh .pl2 file.

What does this mean?
It seems obvious that Shredder has a bug which relates to the number of lines
being analyzed. For some strange reason it does a better job when more than one
line is being analyzed, at least in this position! (I haven't tested any other
positions.)

Moral of the story:
If you are looking for the truth in a position with Shredder, be sure to
increase the number of lines of analysis!
:)
WP



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.