Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Why should I use particular books in engine tournaments?
- - By Overlord (**) [it] Date 2009-07-03 13:58
Why?

If I want to test engines and make them play tournaments with random openings (btw... is this a good idea?) what's the difference if I use, for example, the Fritz11.ctg or any other book?
I think that (apart from Fritz11 in this particular case) all the others would be treated in the same way... am I right?

Why do such a great number of books exist? free and private?
Why do people "test" books? (and above all, what does exactly mean "test books"?)

Thanks for your replies guys ;)
Parent - - By Uly (Gold) [mx] Date 2009-07-03 17:07

> all the others would be treated in the same way...


No, neutral books are better for the cause because the other ones may be tuned specifically for an engine or a style of play, so, you could give Zappa a book that goes with its playing style and Rybka a book that goes against her playing style, results would be skewed in Zappa's favor (Rybka would still win, unless the books were extreme) a similar situation happens to all the other engines.

Books of this epoch are specially tuned for Rybka 3 (because she's the strongest chess entity around), but they could perform very differently for Rybka 4, perhaps, 4th place book would jump to the top, or a small broad book would beat them all because Rybka 4 finds better moves on her own than current theory,  (the effect of a bookmaker refuting a book, in real time) etc.

Basically, the book is a huge factor, so you want to nullify it to know the engines' strength.
Parent - - By Overlord (**) [it] Date 2009-07-03 18:36 Edited 2009-07-03 18:40
hmmmmm.... so all the efforts that I see around here to test books are bounded to find "neutral books"?
Does it take so much?
And so why some books are "private"? Why someone doesn't want to share the fact that ha has found that the position after "7. Axc6" (for example) is a balanced position.
After all "a book" (a ctg file, I mean) is a subset of what we can find in "opening theory books" written on paper. Isn't it?

Can you give me some link to threads where good books can be downloaded? and especially where I can learn something about this subject that you are explaining to me for (my) first time?

Thanks
Parent - By Jonas (****) [at] Date 2009-07-04 00:37
well, reason is they play against each other on a server. they try to have the best book
some authors make books for the public or they create them for different engines and some are making private books to win games on the server.
Parent - By yanquis1972 (****) [us] Date 2009-07-04 00:58
this is what i've downloaded for use in engine matches -- http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=11444

the perfect 2009 book -- it's apparently 12 moves deep, based only on human-human games, & balanced.  so you get to an early middlegame position that follows strong & recent theory & is roughly equal. that's the idea anyway, i haven't used it yet, but it seems like the best bet for reliable & non-repetitive engine-engine games.
Parent - By Uly (Gold) [mx] Date 2009-07-04 03:37

> hmmmmm.... so all the efforts that I see around here to test books are bounded to find "neutral books"?


No, I was replying to the "if I want to test engines", in where performance books (see Permanent Brain's post) are going to be biased.

> And so why some books are "private"?


I think the mian point of performance books is for people that was to create something that is the best out there, Vas reached this with the Rybka engine and now programming an engine as strong as Rybka is really hard, so people create opening books, in where it's easier to be on top. Public books suffer from the flaw that all people can download them and tune their own books against them, so by keeping them private you protect your work from being beaten.

>Why someone doesn't want to share the fact that ha has found that the position after "7. Axc6" (for example) is a balanced position.


Because then other people can add that line to their books and the advantage of having such information disappears. But eventually, if the tournament games are released, people are going to add that line into their books.

I don't have links to the best books handy, sorry, but there's usually links pointing to them on this subforum, look around.
Parent - - By Permanent Brain (*****) Date 2009-07-04 01:03 Edited 2009-07-04 01:06

> what does exactly mean "test books"?


The intention, or purpose of several types of books is different:

1. Test books: Books made for testing the strength of many different engines by big numbers of engine vs. engine games. These books must not favour a particular engine by specific preparation. Also, typically their depth is medium deep or is limited e.g. to 12 full moves, to have engines starting to calculate in early middle game. I think that reflects also in which types of positions an engine will be used for analysis, in a chess player's practice. The tester groups like CCRL and CEGT try to create standardized, more or less clearly defined conditions at a kind of "laboratory standard", to get reliable test results. Neutral test books are part of these configurations. Alternatively, also sets of predefined opening variations are in use.

1a., In contradiction to that, but traditionally the SSDF tests with engine specific books (usually as provided by the commercial publishers).

2. Books for Playchess' machine room: I call them "performance books." These books are intended to score as good as possible, typically combined with Rybka 3 (but sometimes also with other top engines), in a kind of online engine sport competition which permanently happens there 24/7. It seems to be an endless battle between novelties and their refutations, often in sharp Sicilian Najdorf variations, which seems to require close observation of "daily" :-) developements and frequent adjustments. In contrast to the environment test groups provide, the situation at Playchess is chaotic (different hardware, different settings, different books...).

3. Opening References: Large books created from good collections of human master games, representing the opening practice of Grandmasters and other strong players. May be enhanced with "unplayed" theory from analyses, as published in opening theory sources.

The engine specific, large commercial books which are sold with, or for commercial engines (Rybka, Shredder, Fritz etc.) need to contain elements of (2) and (3). I would say (3) is the much more important aspect for a software company, because customers dedicated to the engine sport as described in (2), certainly have only a very small share. Even though the forum may give the opposite impression, sometimes.
Parent - By Overlord (**) [it] Date 2009-07-04 05:20
Thanks Permanent, thanks vytron, thanks Jonas for your replies and thanks yanquis for the link you posted.
Now I understand a little more of this subject and I will try to run a Perfect2009 eng tournament to see what chamges in respect of the "Fritz11.ctg eng-eng tourn" that I tried.
I'm going also to search other threads on this subject on this Forum.

Cheers ;)
Up Topic The Rybka Lounge / Computer Chess / Why should I use particular books in engine tournaments?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill