Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Shredder's bug?

Author: William Penn

Date: 06:46:17 12/21/03

Go up one level in this thread


On December 21, 2003 at 08:53:39, Jaime Benito de Valle Ruiz wrote:

>I've never read any thread about this before, so forgive me if this is a
>recurring topic.
>Many people have noticed that the Principal Variation shown by Shredder is
>sometimes ridiculous, to the point of giving out its queen and a rook for
>nothing, or allowing a silly mate. If this is “really” what Shredder is
>thinking, then I’m truly surprised that it’s such a strong engine… considering
>that it chooses absurd lines that lead to self-destruction. My engine sometimes
>does something similar, and I’m sure it’s a bug, because the final static
>position of the PV does not match the score is showing anyway (I have other
>things to correct before the PV at the moment anyway), and neither does for
>Shredder.
>Is this a really a bug, or just a trick to conceal the way engine is doing the
>internal search? Because I don’t believe that those PVs are the ones Shredder is
>using for analysis, and still play decent chess.
>What do you think?
>
>  Jaime Benito

I'm one of those who has noticed this bug in the Shredder 704 engine. That's the
only engine I use. I don't know about the Shredder 7 engine. It is a terrible
bug because analysis can never be trusted beyond the first move. That becomes
obvious when you analyze for an hour or more in infinite analysis mode, such as
during a correspondence game in progress. It is true that the first part of the
analysis, the first few ply, are usually reasonable moves, but not always.
Sometimes the nonsense begins only 2-3 ply deep. It is clear that those
positions are NOT being tested for checkmate, because "mate in 1" is overlooked
frequently. Giving away the queen is also very frequent, or overlooking a free
capture of the queen. These nonsense variations often contain kamikaze
sacrifices (sacrifices without a purpose).

My tests in real correspondence games versus masters shows that the Shredder 704
engine is very strong, perhaps near Elo 2450 on the correspondence chess Elo
scale (which is probably about 300 Elo points lower than the over-the-board Elo
scale). So I conclude that the first move in the analysis by Shredder 704 is
strong and reliable. It is only the subsequent moves that are suspect.

Why?

I don't think we can know the reason why - unless the engine's author wishes to
comment. That would be very interesting! I suspect that he has simply not
considered that analysis is important, and has sacrificed everything to produce
the best first move.

I certainly hope this will be corrected in Shredder 8, reportedly to be released
on about January 8, 2004.
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.