Computer Chess Club Archives


Search

Terms

Messages

Subject: Shredder and Engine Book Learning

Author: Stephen Ham

Date: 12:47:32 04/24/03

Go up one level in this thread


On April 24, 2003 at 12:28:36, stuart taylor wrote:

>On April 24, 2003 at 11:46:06, Stephen Ham wrote:
>
>>On April 24, 2003 at 09:45:38, Kurt Utzinger wrote:
>>
>>>On April 24, 2003 at 09:33:36, Eran wrote:
>>>
>>>>If yes, what are they?
>>>>
>>>>Eran
>>>
>>>    Shredder 7.04 is a fine software, perhaps the strongest chess program
>>>    at the moment ... but it's sometimes too optimistic
>>>    Kurt
>>
>>Kurt is correct, per usual. Shredder 7 and Shredder 7.04's evaluation function
>>is too optimistic. Nonetheless, Shredder 7.04 is probably the strongest engine
>>on the market. I do have a gripe too about its opening book in the ChessBase
>>GUI, since it's not nearly restrictive enough. I've manually corrected this on
>>my Schredder, so it's very strong (it now plays only MY opening lines that I
>>play in Correspondence Chess). Sandro Necchi, the author of Shredder's Book is
>>correct that the Shredder Classic GUI restricts its openings to only the
>>strongest lines.
>>
>>I also believe that version 7.04 is clearly stronger than 7. This is especially
>>evident as the time-control increases.
>>
>>As for Shredder's strengths/weaknesses, here's my review, found at
>>ChessCafe.com:
>>
>>http://www.chesscafe.com/text/review365.pdf
>>
>>All the best,
>>
>>Stephen
>
>I've just spent a long time reading your review of Shredder 7. Have you noticed
>Shredder 7.04 correcting many of those weaknesses of Shredder 7?
> In all your praise for Shredder 7, you had it only very vaguely that it might
>be a little stronger than Fritz 7, whereas Fritz 8008 is clearly reported as
>being quite a bit stronger than that.
> Does the upgrade of Shredder 7 to 7.04 make all the difference, to overtake
>Fritz 8008 as well?
>S.Taylor

Hi Stuart!

Shredder 7.04 hasn't corrected all of the weaknesses seen in version 7, but it
is stronger. As for Fritz 8, I confess that I'm not qualified to answer this
since I don't have Fritz 8, just Fritz 7. Now that I have S 7.04, I can say that
it is clearly stronger than both S 7 and Fritz 7, based upon my tests over
various time controls. I believe that the real differences are best seen at
longer time controls. I think too that the opening books have a great deal to to
do with success or failure.

In the ChessBase GUI, for some reason, some of the best moves in Shredder's book
have a "?" next to them, and are color coded not to be played. Sometimes, the
move is forced, but the coding in the book forces Shredder to waste several
minutes on obviously forced moves. Therefore I've manually fixed this at every
opportunity. Still, it had cost Shredder a lot of points in my test games to
date.

As stated above, I've manually modified Shredder's opening books to play the
lines that I play, resulting in even more success. This may be an unfair
advantage. To compensate, I've tried to improve the Fritz book to offer it
better counters to my lines. I just bought Chess Tiger 15, and am gradually
modifying its opening book too (although it's book doesn't seem to have the
errors that Shredder had).

Here's one final thought of mine. First of all, I need to state for the record
that I'm not an engine expert. I'm about as ignorant as one can be on this
topic. Still, I'm not a fan of "Book Learning" at all, in spite of what
Christophe Theron wrote in its favor. When I allowed my engines to play in
matches with Book Learning on, they sometimes convinced themselves not to play
the best lines, just because of one bad result.

Let's create a hypothetical situation for clarification. Let's say that a
certain line is busted and that an engine playing the refutaion will win
99-times versus 1-loss. But if that 1-loss was in the very first game, the Book
Learning function would weigh so heavily against repeating that "best" line that
it would subsequently always play an inferior line, in spite of the fact that it
would otherwise win the next 99-consecutive games in the "best" line.

Another problem is that the reason for the loss/win may have nothing to do with
the opening. Nonetheless, I believe that Book Learning will label the opening
based upon the result of the game. I could be wrong here, but this is just my
perception. I've seen many games where the stronger engine gets a bad opening
against a weaker engine. However, the stronger engine will often still win that
game because it's a stronger engine. Even though both engines evaluated the
opening as more favorable for the weaker engine, I've seen "Book Learning" stear
the stronger engine back into that bad opening again, because of the previous
win. Also, the weaker engine's book also avoids the superior opening for the
same reason. That's a real problem!

For these reasons, I no longer allow "Book Learning." After each game, I'll play
through each line and try to improve each engine's book manually, if needed. I
think that the net effect of my testing is that I've improved the opening books
of all my engines significantly. They now have better quality books (play with
more precision) and also deeper books. I don't try to tune the books to the
engine; instead I try to give them what I believe to be objectively best. As a
consequence, I now see them playing much better chess.

But that's just my perception.

All the best,

Stephen



This page took 0.03 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.