The scaling with time of opening books

[Thomas Zipproth]

Hi Thomas
When are you going to release the Commercial version ?
Right now every Tom, Dick and Harry is using your excellent Cerebellum book, which you make with so much effort, for free....gives me no advantage.
People who are prepared to pay money, should get an advantage over those who want only free gifts !

Thanks for your support.

The release of the Commercial version will be most likely late November, because there is a complex Gui involved which requires some testing. All the single parts are ready and must now be tied together. I will try to release a Beta Version as soon as possible. You are right, updating the book nearly every day is a lot of work. When the Commercial version is released, everyone can do that for himself, and also a group of people can work together in updating the book. So after the release there is no more need to make daily updates.

I will not stop completely with Brainfish - Cerebellum, but I will stop monitoring the daily Engine Games so close and update the Brainfish book not so regular. From this time on the Cerebellum books created with the Commercial Version will usuallly be stronger, because more regular updated, containing more analysed games and not released immediately to the public. And for some time I will focus on including Grandmaster games, because Cerebellum is also intended to be used by professionals for example in opening preparation.

Thomas

[Guenther]

I am still working on it. Progress is slow and I hope I have not taken on too much..

https://docs.google.com/spreadsheets/d/ ... o7o87dnjvs

[Laskos]

I am still working on it. Progress is slow and I hope I have not taken on too much..

https://docs.google.com/spreadsheets/d/ ... o7o87dnjvs

Very interesting. It seems Andscacs benefits even more from the book. I was somehow under impression that SF would benefit more, because I suspected the books are tuned for SF At 20+ increment our results are close. Waiting for other data, thanks!

[Guenther]

I am still working on it. Progress is slow and I hope I have not taken on too much..

https://docs.google.com/spreadsheets/d/ ... o7o87dnjvs

Very interesting. It seems Andscacs benefits even more from the book. I was somehow under impression that SF would benefit more, because I suspected the books are tuned for SF At 20+ increment our results are close. Waiting for other data, thanks!

Interesting yes. I was not content with the initial test setup and wanted to see how different programs would do here.
The disadvantage is that it will take very long to complete the current test with all desired data... (and I sure cannot go above 240+3)
Probably I will add another column with average bookout depth later.
The PGN files will be available too for further data mining ;-)

I noticed also a huge discrepancy in draw rates! Is it possible that engines have an immanent 'fingerprint' for this, independent from books/time control?

Aother explanation would be increasing strength, but Andscacs is already one of the strongest programs and in the top 10.
Main difference is that it cannot use tablebases, but from a first look it does not seem the reason for that low draw rate.

On the other hand I think I have seen more sudden eval changes in the middlegame, at least in some games I watched, which led to decisive results.

[cdani]

Thanks for the test! Is really interesting. I find difficult to interpret the results.

Andscacs benefits more of the book because it plays worst than Stockfish and needs more help to leave the opening in good conditions? But as it plays weaker one can expect that it loses easier than Stockfish the advantage of the opening. Or maybe this is related to Andscacs being not able to hold well bad positions.

[Ozymandias]

Thanks for the test! Is really interesting. I find difficult to interpret the results.

I'd say it plays worse at the very beginning (depth 3-15), and after that, it behaves more like SF.
As seen in other scaling tests, the more time you give it, Andy strengthens and those flaws in the early opening are more easily taken advantage of.