FEOBOS v10 available ... first good version!!

[Frank Quisinsky]

Hi Jeroen,

OK ...

Much more interesting to read for us is a comment like:
So many years computer chess do nothing for openings / opening books, for new opening book features in chess GUIs. Allways the same boring material.

Main interest is to create a tool for the future. The Excel files are on high end level because all what is possible to use from *.epd is included with a lot of new ideas in Excel. Excel is perfect for such things.

Normely I await from opening experts more a thanks. Because from our experts I see nothing what is new and interesting since many years. So I am thinking OK Frank ... made it better. I can't make it better but our engines can do that.

Now to your comments:

1. Very short one-move-book-exit analysis.
The time for one analysis is factor x20 in comparing to CCRL 40/40 with 10 threads on i7 4.3GHz and 60 seconds per analysis. Possible that in 10 years an available Rating list have the same conditions we are using for FEOBOS. For the moment is the "Very short one-move-book-exit analysis" more as enough for that what we do in engine-engine testing.

Example:
If I would say ... 41.000 positions with 10 engines = 410.000 analysis x one minutes ... I need 1 years (what we do). With two minutes for each analysis 2 years, 3 years for 3 minutes. Very expensive to do that. And believe me for perhaps 2% better results ... made no sense.

But the deal is: Users can create the analysis with 5 minutes on faster Hardware or what they like. Very easy to added the *.epd in our Excel files. All Excel and basic files to FEOBOS project are open no problem. We have here no secrets.

In 20 years with 100 cores and three Minute per analysis ... we can do that and ... the Excel file is developed and ready for it. The main work is complete ready for future events. I like that ... I can used with newer Hardware and engine the Excel file and can start FEOBOS again. The results will be again a stronger book.

For the moment is the FEOBOS qualitiy of analysis for a rating list system, engine tourneys on a very high level.

2. Engine authors
Can looking in Excel and can sort the results with eval information. Can directly see where the own program have Problems and can try to create to fix that. One months a big Topic for Klaus and myself. We developed at first all what can be important for engine authors.

If I read your message step by step ... I am sure ... you have never looked in our Excel files in detail. I create a dia-show, you can looking what we do in the graphics from the worksheets (dia-show).

3. Balanced moves.
What is a bad and what is a balanced move? All that with all the different evals by 10 different engines we are unsing (very complicated but possible to solve) ... pure knowledge after my work on FCP live book. Means my settings are pure knowledge because I checked most of the 300.000 FCP Rating List games by hand for handtuning in FCP Live Book over three years in combination with strong written books to the topics openings. At the moment I have around 370 books only to openings and I can looking if I am sure with my own chess knowledge.

But we are humans and around 1.000 - 1.400 Elo weaker as the engines we are using for FEOBOS with the time control we are using.

But if you have a problem with an "cut eval ... example 0.70 / - 0.40 for Houdini 5 ... no problem. Again, we developed the Excel file very open.

You can set the setting you like ...


So you can added your own settings and Excel calculate the test-set and calculate the ranking position number with your own setting.

Maybe you find the time to study our ideas and maybe you have time to give us hints for improvements, because you are one of the opening experts we have. We have an open ear and maybe we can make FEOBOS with your helps better.

For the moment 10 of the strongest engines create the book in teamwork. The final results are great for eng-eng constellations. And honestly ... I can see nothing what is better ... because what human do have to many errors ... means if a human try to create such an opening book to 500 ECO codes ... not possible in my opinion ... not with such a high qualtity.

So I give my thanks to the programmer of engines. Without the strong engines today are available ... never FEOBOS can produced such a strong results.

I do nothing ...
Klaus do nothing ...

What we do is ... we make that what engines do readable for humans with stats. This with a lot of complete new ideas to our opening Systems and features for the future of computer chess.

Sorry for my bad English.
Have today not a good day.
Need a new car ... the old is defect.
I like the car.
Sorry if my message are to aggressive for you.

Best
Frank

[Gusev]

In the Contempt 1 .ctg book, how come after 1. e4 e5 2. Nf3 Nc6 3. Bc4 Be7 the move 4. d4 is not an option?! Only 4. O-O and 4. d3 are considered.

[Frank Quisinsky]

Hi Dmitri,

6 positions in upd.1 files with 15.468 positions (can be found on FEOBOS site, basic databases, 6.27 Mb).
upd.1 databases will be checked in the next 4 months from all 10 engines standard engines too.

In upd. 1 database I collect the lines, not included in the first 26.146 positions ... the missed lines.

In around 4 months the upd.1 database will be checked by the same 10 engines too and FEOBOS project is over / ready.

Best
Frank

[Frank Quisinsky]

Hi there,

I am looking in the games from the still running Wasp 2.5 test-run with FEOBOS v10 Contempt 4.

Code: Select all

Games        :    876 (finished)

White Wins   :    276 (31.5 %)
Black Wins   :    162 (18.5 %)
Draws        :    438 (50.0 %)
Unfinished   :      0

White Perf.  : 56.5 %
Black Perf.  : 43.5 %

ECO A =    127 Games (14.5 %)
ECO B =    240 Games (27.4 %)
ECO C =    152 Games (17.4 %)
ECO D =    132 Games (15.1 %)
ECO E =    225 Games (25.7 %)

Now we can thinking ... 25.7% to ECO E00-E99 is to high. But this is right for a book without move transpositions. Also the 15.1% to D00-D99 is absolutely right. With move transpositions ... ~ 19% to E00-E99 and clearly more to D00-D99. ECO code stats to A00-A99, B00-B99 and C00-C99 is the same as in TOP GM theory. Really wonderful what I can can see here.

Point 1: ECO allocation between the ECO Codes are wonderful for FEOBOS v10 with Contempt 4.

Next point 2:
The shortest draw game from the first 876 games = ended with 23 moves. That is to 100% perfect. I am thinking before that around 0.5-1% games ended with draw undo move number 20 and around 3% with draw undo move 23. I am very happy about it.

0 games !!!
Here Contempt 4 book settings works great!

Next point 3:
No bad lines are in the 876 games!
But this is clear because 10 TOP engines analyzed the basic database.

Next point 4:
Draw quote from Wasp with around 2.840 Elo vs. opponents with an average from 2.850 Elo is 50%. That is perfect too. Normely around 58%.

FEOBOS v10 with Contempt 4 is working wonderful.
I am very happy with the results I have. So all the work around FEOBOS is right and I can really create my dream book.

Next point 5:
No doubles games in the matches with only around 16.000 lines I am using from the 26.000 basic lines with Contempt 4.

Bad point 1:
The book with move transposition can be available end of January 2018. Sure that with Contempt 2 the results will be great too.

It will be very difficult for an other current available opening book to create better stats for hardcore engine-engine testing. And I can be sure that I create with FEOBOS better Elo statistics (for a rating list system).

And to Wasp:
A lot of fast won games, but one game Wasp lost vs. Cheng very fast. I can't see a big mistake Wasp made here. For the others: Wasp is really very strong in the beginning of the games, produced many fast won games and rarely a fast lost games ... also vs. a stronger group of opponents.

Here the database for checking how strong FEOBOS v10 Contempt 4 do the job:

http://www.amateurschach.de/download/wa ... _games.zip (1.21 Mb)

A good day ... much better results as I await, near to the perfection for myself.

Best
Frank