Computer Chess Club Archives


Search

Terms

Messages

Subject: Regarding Crafty Opening book,Review by Mr.Peter Berger

Author: Madhavan

Date: 07:33:54 07/13/05


Hello all,

I'd like to move the post from CTF to here because the topic was completely
about opening book,brilliantly written by Peter Berger.

This stuff could raise discussion or atleast something to read about Peter's
view on how he made the book.

Here is the actual link being posted in CTF forum,don't view the whole thread
please :-)

http://www.talkchess.com/forums/2/message.html?175013

I don't have any idea about how to make opening book for engines,so please Peter
is the right person to answer any of your questions.

May be I could paste the stuff carefully ignoring his stuff about Vincent :)

[Quote]

It might also
matter that I used to do tournament books for another amateur program that was
often underestimated in the past , but used to do well with them in the CCTs,
Gerrit Reubold's Bringer ( on these I actually spent a few thousand hours LOL).

Another rumour I mostly don't really believe in is that books have to be
carefully adapted to the engine's character, at least not for the strongest
competitors, because their style is not really that different anyway. I don't
think that if they had to swap their preffered openings, but still could have
the same level of book quality, you would note any difference in results. But
then they have all been playing nothing but the Najdorf for years anyway :) .

The same is not really true for Crafty. It has some clear strengths but also
some clear weaknesses that have to be taken into account when you want to do a
book. And when you think of the tournament games , the first two games against
Shredder and Sjeng basically had no middlegame at all ( which was Shredder's
decision in game 1 though ;) ). I think this is mostly what his positive
feedback is based on, besides that he thought that it was a very clever idea to
be out of book against Diep after 1. d4 d5 2. c4 e6 3. cxd5 , which looks a bit
superstitious to me :) .
No one (except me) has any additional data to reach any sensible conclusion
about the quality of Crafty's book in Ramat-Gan, other than the 11 games played
in the tournament, so I'd say people just have no idea, it's nothing but
rumours.

The book is not availlable for download yet for practical reasons only. A
considerable time of preparation goes into lines no one ever plays ( say how to
counter 1. b4 or 1. f4 or 1. g4 or other stuff like that) . This part of
preparation doesn't have to be redone even if you change the whole repertoire,
but it could make Crafty vulnerable if it were published.

The major reason why I think that it is easy to considerably weaken an engine by
an opening book is the strength explosion of engines in recent years (there
would be some other reasons too). There are a lot of tactical mistakes in human
analysis, that you don't want to repeat when the alternative is a move an engine
has calculated for itself for about 5 minutes on sth like a Quad Opteron.

I think this is difficult to admit for semi-professional book cooks : two games
from Ramat-Gan that I think are worth a closer look are Crafty-Shredder and
Falcon-Shredder e.g.

Another issue would be  the kind of stuff you get with automated books when you
find the engine in a Ben-Oni or French structure that doesn't suit it at all. It
often would have been better to have no book at all then :) .

So it looks like a sensible approach to me to at first make *very* sure that you
don't weaken the engine before getting overly ambitious :) .

When it is about testing: I basically have no clue if "my" book is really any
better than a random book in an absolute sense ( whatever this is supposed to
mean) - isn't that a scary thought :) ? I don't think so , as for one it is very
difficult to setup a valid test scenario (learning is an issue e.g.) and on the
other hand the time saved can be used in a better way IMHO :), e.g. by ironing
out known weaknesses.

Oh - and btw: I haven't written a single line of the new one yet and won't start
for another two weeks ( very amateur-like) . But Chris Taylor's autoplayers will
support the Crafty team again, and they produce data faster than anyone can draw
any conclusions from anyway. Testing is mostly done against random top
commercials at medium time control.

[End of Quote]

I'm weak at openings,Peter.They are the audience,I'm not the right person to
answer the above master piece.

Thanks



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.