Opening DataBooks

[Anonymous]

I wish to introduce a topic on Openings DataBooks
There are currenyly a lot of yhem , and everybody claims " my book is the best"
But the best for WHAT?
Is it possible to throw some clarity on this debate?
There are many uses of an ODB , and none can be the best for all uses
So it wold be interesting :
1) To classify the different uses
2) For each use , to define the requirement of a good ODB
As for me , I shall propose 3 classes :
Working Books , associated to a Database ; They may be general or specialized
"Common " Books for tournaments between Engines
Optimized Engine books
If you agree , there is left to define the requirements for each class

[Kirill Kryukov]

What's "Openings DataBooks" ? Is it the same thing that most people call "Opening Books", or is it something different?

If you agree , there is left to define the requirements for each class

Well, engine-specific books certainly have only one requirement - to make the engine as strong as possible. It's done by avoiding positions the engine does not understand, and by including lines that lead to positions where the engine can play well. Commertial engines like Fritz or Shredder have books optimized by a specifically hired professional chessplayers, AFAIK.

Books for engine tournaments usually have purpose of fair comparison of engines. They should have big variety of positions, while avoiding exchanging too much material or developing too far in general. Some good books for engine comparison are limited to 8, 10 or 12 moves.

[Edit]

There are currenyly a lot of yhem , and everybody claims " my book is the best"
But the best for WHAT?
Is it possible to throw some clarity on this debate?

In my experience, most people who make opening books know very well for what purpose they are doing it, and also most of them don't claim "my book is best". Well, that's only my experience of course.

[Anonymous]

Hi Kyril!
You define very clearly the requirements for each class , but things are not so simple:
At what extent are they fulfilled?
Are there conflicts between some of these requirements?
I shall give just 2 examples :
Are the books of commercial engines really optimized? I doubt it
Junior engines are all very strong in tactics , and yet , with Black against Ruy Lopez , their books make them play almost exclusively Closed Defence , of which the slow maoeuvring style doesn't suit them ,although they play Open Defence much better ( maybe GM Alterman doesn't like it)
All the same , with white in Open Games , they should play 3 Bc4 rather than Bb5
2) In Common Books
The wish of including only balanced lines leads to suppress the most interesting ones (because nobody agrees on their assessment )
Hence , with these "balanced" books engines games are often very flat ,
and not selective enough for the best engines
I made a try : I get ride of this requirement of balance , and I kept only lines that give interesting game
So , I got a very short book (250 lines ,but I intend to grow it up to 1000 )
That proved very selective , and at end as fair as usual common books
Friendly Yours
Claude Le Page

[Kirill Kryukov]

Hi Cladde!

I don't know what the question really is... For engine-specific books there is of course a conflict between a book quality and the resources (time and money) spent on book tuning. Hire a team of top players, give them money and a year of time, and that engine will win most of games during the openinig. I don't have comment about Deep Junior book versus Ruy Lopez, it's just a matter of developers' preferences, talent, and time spent to tune the book. I guess developers may have their own idea what suits and what does not suit their engine...

Are the books of commercial engines really optimized?

They are optimized as much as they were able to do it. They surely did not have any other goal developing a book.

The wish of including only balanced lines leads to suppress the most interesting ones (because nobody agrees on their assessment )

Well. The question here is how you define balanced lines, and how you define interesting ones. To me the line is balanced (in engine-engine context only) if after many games the white score is not too big, and not too little for that line. To me the line is interesting (also in engine-engine context only) if it is balanced, and the draw rate is relatively small for that line, also after many games. It so happens, that interesting lines are not long, usually not longer than 12 moves. This is good, because engines have more chance to think and show themselves in middle-game, where their strategic abilities are most important.

Hence , with these "balanced" books engines games are often very flat ,
and not selective enough for the best engines

What is flat? Does it mean the game balance does not change too much? If the opening leads to too high draw rate, it means the opening is not interesting, most usually because it's too long. Most openinig that are about 10 moves result in interesting games.

OK, one objective criteria of interesting game can be a ponder hit rate - a percentage of expected moves played. I'd say a game is interesting if the ponder hit is small, and uninteresting if all moves are expected. In most engine-engine games, where opponents have comparable strength, the ponder hit will be between 50 and 60%, which means there are 40-50% of unexpected moves (under long time control). Such games are usually intersting. If the ponder hit goes to 70% or above, I'd say the opening is not good, or one engine is a clone of another.

I made a try : I get ride of this requirement of balance , and I kept only lines that give interesting game
So , I got a very short book (250 lines ,but I intend to grow it up to 1000 )
That proved very selective , and at end as fair as usual common books

1000 lines is too few. Try 250 games with it and you'll get duplicates. The main purpose of a book is to avoid duplicate games, otherwise the engines would simply play from the starting position. (In engine comparison study context). One good way to go is to take a huge book of human games, and limit the opening length to somewhere about 10 moves. There will still be unbalanced openinigs, but such games can be filtered out in post-processing. (It's my theory since I did not make a tool for it yet). But the opening variety will be huge which is always good.

BTW, did you try Sedat's books? Recently I use them for engine-engine games and most games are hard-fought.