Book testing (tuning) with ProDeo

ProDeo 3.1x is a special version with the purpose to tune (analyzed) Polyglot opening books by playing matches at rapid time control. Opening books are never perfect but in a convenient way it's possible to weed out most bad or not so good moves. In the below examples we pitch two equal Polyglot books against each other, book-1 the normal way, book-2 the smart way. The smart way does the following things:

1. The first step checks the score of chosen book move with the score of the previous book move and if the score drops with >= 1.00 that book move won't be played. And if that was the only move available from the book the engine starts to calculate ignoring the book move.

2. The second step investigates the cases when there are more than one book moves available. If the score of the chosen book move is lower than >= 0.50 than the score of the best move, the chosen move is replaced with the move with the best score and finally the new chosen book move is compared with the previous book move -- as above -- but now with the 0.50 margin.

3. Book decisions [1] and [2] are stored in the file book-remark.txt as a guide to make modifications for instance with SCID. Results:

....

http://rebel13.nl/prodeo/prodeo-3.1.html#book

I'm trying out the book testing of Prodeo 3.1x
I used the including ProDeo.bin books, and I used the standard settings.
I just changed -rounds to 20
The text file 'book-remarks.txt' is not written so something goes wrong. Maybe it is because 20 rounds are not enough?  
I don't get any error messages. Here is a screenshot:


Where can I change the default  [Blunder Margin [1] = 100] and [Blunder Margin [2] = 50]?

Is it here:


And should the default second score=500 not be score=50? From 50 centi pawns?

Can I use book testing with Prodeo 3.1 x also with analyzed books made with Poly 1.7b? With the option W or B to make only an analyzed book for the white moves or the black moves.

Jonathan003 wrote:

I'm trying out the book testing of Prodeo 3.1x
I used the including ProDeo.bin books, and I used the standard settings.
I just changed -rounds to 20
The text file 'book-remarks.txt' is not written so something goes wrong. Maybe it is because 20 rounds are not enough?  
I don't get any error messages. Here is a screenshot:

As the screenshot shows "'book-remarks.txt" is surely written in the root, it might be empty because there no cases to mention.

Quote :

Where can I change the default  [Blunder Margin [1] = 100] and [Blunder Margin [2] = 50]?

See "book.ini" in the root.

Quote :

Is it here:


And should the default second score=500 not be score=50? From 50 centi pawns?

These values have nothing to do with the "Blunder Margin" settings, again, see "book.ini".

Quote :

Can I use book testing with Prodeo 3.1 x also with analyzed books made with Poly 1.7b? With the option W or B to make only an analyzed book for the white moves or the black moves.

Sure.

Thanks for the explanation.
Can you give some guidance how many rounds I should run for a good result?
I use a small analyzed bin book made with poly 1.7b with the option W to only analyze the white moves. After wards I made an only black book from the same pgn with polyglot-tolerant and a merged the white and the black books.
I had to add the black moves before running matches with ProDeo 3.1x, otherwise black went out of book soon.

If I use the option polyglot_tolerant info-book [-bin inputfile] [-exact]

I get this information for the ProDeo.bin book:

Lines for white : 36831
Lines for black : 33767
Positions on lines for white : 452548
Positions on lines for black : 413199

And I get this information for the white_rep_lo-pref_40and0.bin book I use.

Lines for white : 2926
Lines for black : 3745
Positions on lines for white : 6337
Positions on lines for black : 5671
Unreachable white positions(?) : 0
Unreachable black positions(?) : 0
Isolated positions : 0

If the default for numbers of rounds is 1000 games for the ProDeo.bin book, what would be a good number of rounds for the book white_rep_lo-pref_40and0.bin?
And is there some rule I can use for other books?

I understand you test with only 4 threads and that's of course a problem to play 1000 games. But you can increase that number. Edit 'start.bat' -

ccc -concurrency 4

to

ccc -concurrency 8

And things go twice as fast in case your PC has 8 threads.

Secondly, start with 100 games. What counts are the suspect cases reported in the 'book-remarks.txt' file.

Thirdly, and this is an overall answer (advice) also on your PM. It's impossible to create the perfect book solely based on computer analysis because of the human preference for openings. There always has to be the 'final touch' by the human hand based on his likes and dislikes for opening theory.

What is possible is to take out blunders and suspect moves based on computer scores. The ProDeo Book Testing is just one of the possibilities, POLY offers you more, IMO the [F5] function is the best, see - http://rebel13.nl/text/result.txt

With all these informations manually tune your books with SCID and re-run the match for a better result. It's the way Jeroen Noomen created the strong REBEL books in the 80's and 90's, the final human touch.

Another way is to create opening books from eng-eng matches between 2 strong engines. There likely are no blunders. The disadvantage is that books lack variation but if that's no problem for you try that path.

I agree the human should always have the last hand in deciding what variations to include. But there should be better tools for creating repertoires for otb games imo.
But it appears not many programmers are interested in making such tools. To do what I want I need six different software and tools.  
Fritz 17 has included standard repertoires and Chessbase 16 has some functions to automatically create opening surveys. It is a first step but it is not sufficient for easily making repertoires. My idea is to let an engine create a repertoire overnight by comparing a big reference database with a database of games of a player who's playing style you like (it can be your own games), end checking the analysis.