FEOBOS v10 available ... first good version!!

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Frank Quisinsky
Posts: 4852
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

FEOBOS v10 available ... first good version!!

Post by Frank Quisinsky » Tue Sep 26, 2017 9:56 am

Hi there,

now all 10 standard engines analyzed 26.146 positions with 10 threads and 1 minute per position. As next important forgotten lines are to analyse with the same group of engines (15.468 positions). The final FEOBOS version will be version 20 around middle - end of January 2018.

Komodo 11.22 x64 as last engine in KECR on ranking position 5. Without contempt and over 10% 0,00 evals also very high. Not so high as ASM Fish with over 15%. New TOP engine for opening analyses is Houdini 5! Sure that Houdini 6 is again stronger here. Andscacs on second!!

We can speaking from the final results for the first 26.146 basic positions. We have over 12.000 position 0 of 10 engines gave an 0,00 eval. We have one position with 9 different best moves (an E11 line). 4 positions with 8 different "best moves" etc..

The complete database is sorted with a ranking system. Very easy to create any test-set you like to ECO codes you like.

You can find 5 opening books in download directory (all without move transpositions). Contempt 1 - 5 books in Polyglot *.bin and in *.bkt & *.ctg format.

Have a look here ...
Image

We have 22.145 of 26.146 good positions (without bad moves). The 22.145 moves are sorted in 5 "Contempt groups".

KECR and FEOBOS Excel files are updated too.
Also the DIA-Show is updated with all current results.

FEOBOS:
http://www.amateurschach.de/main/_new-opening-book.htm

Best
Frank

PS: The FEOBOS book with move transpostions and Contempt 2 will be available with the final version middle - end of January 2018. For the next 4 months no FEOBOS book updates will be available, to complicated to create it with our Excel files during the time 10 engines have to analyse the upd.1 database. But with FEOBOS v10 you have the final version from the first 26.146 database.

FEOBOS is the first opening book with Contempt features.
FEOBOS is the first opening book with an ranking system for all positions inlcuded in the book.
FEOBOS is the first opening book 10 engines analyzed the complete positions.
FEOBOS is the first system you can create very easy the test-set you like to each of the 500 ECO codes for for Rating List systems.

And FEOBOS is the first opening book tuned by engines itself in teamwork.
I like computer chess!

Frank Quisinsky
Posts: 4852
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

Re: Work on KECR ended ... with a lot of new features!

Post by Frank Quisinsky » Tue Sep 26, 2017 10:22 am

Hi there,

in the last weeks Klaus Wlotzka is working hard on the second Excel workbook (first is the workbook for FEOBOS) ... our KECR ratings based on the first moves after opening book moves!

- Elo feature added.
- Comparsion Feature for two engines added.
- Timelife feateure added
- football point system added.

And much more!
Now both Excel files are ready for book optimization.

You can download the work or you can looking for more details in our "Dia-Show".

FEOBOS:
http://www.amateurschach.de/main/_new-opening-book.htm

Best
Frank
I like computer chess!

User avatar
kranium
Posts: 1824
Joined: Thu May 29, 2008 8:43 am

Re: FEOBOS v10 available ... first good version!!

Post by kranium » Tue Sep 26, 2017 6:38 pm

Hi Frank-

Wow...
I took a look at this book and your documentation, and I can't believe how much work you guys must have put into it. Thanks for the effort!

I do have a couple questions:

How does this compare to or differ from cerebellum book?

How do you think it can be best used...can you explain the main purpose (your goal) in a nutshell?

(engine vs engine testing,
online competitions,
correspondence chess?)

PS - How does the spreadsheet fit in?

Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 10:41 am

Re: FEOBOS v10 available ... first good version!!

Post by Lyudmil Tsvetkov » Tue Sep 26, 2017 11:01 pm

is this a book or a rating list?

Gusev
Posts: 1475
Joined: Mon Jan 28, 2013 1:51 pm

Re: FEOBOS v10 available ... first good version!!

Post by Gusev » Wed Sep 27, 2017 12:26 am

How can we modify FEOBOS v10 to make a .ctg that could be used for human game preparation and engine chess at InfinityChess?
kranium wrote:Hi Frank-

Wow...
I took a look at this book and your documentation, and I can't believe how much work you guys must have put into it. Thanks for the effort!

I do have a couple questions:

How does this compare to or differ from cerebellum book?

How do you think it can be best used...can you explain the main purpose (your goal) in a nutshell?

(engine vs engine testing,
online competitions,
correspondence chess?)

PS - How does the spreadsheet fit in?

Frank Quisinsky
Posts: 4852
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

Re: FEOBOS v10 available ... first good version!!

Post by Frank Quisinsky » Wed Sep 27, 2017 4:47 am

Hi Dmitri,

5 x *.ctg books are included in download file.

5 x *.ctg books (Fritz GUI)
5 x *.bkt books (Shredder GUI)
5 x *.bin books (Polyglot)

For Contempt 1 - 5.

All material with readme can be found in the download file with 26Mb).
The download file can be found in download cockpit on FEOBOS site.

To each of the lines you can found in the Excel files the analyzes by 10 engines, or you can download for each of the 10 analyzed engines the *.epd files.

Of course included in 26Mb download file are the *.pgn too. For Contempt 1- 5 5x the *.pgn files and the *.pgn file with all postions (*.pgn file 6).

So you can create for each other GUI the opening book. The *.pgn file is sorted by an own ranking. Each Position have an own ranking, calculated with the Excel file with around 6 millions formulas. If you will have all Information, please download the 26Mb FEOBOS v.10 Version and the Excel files if you will during a match looking what is the opinion by the 10 engines to the position you have on the board.

So, all what you need, not important for which GUI can be found in the 26Mb download file.

Best
Frank

Note:
The available material is ready without move transposition.
The FEOBOS book with move transposition will be ready to the end of the project in around 4 months.
Last edited by Frank Quisinsky on Wed Sep 27, 2017 5:00 am, edited 1 time in total.
I like computer chess!

Frank Quisinsky
Posts: 4852
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

Re: FEOBOS v10 available ... first good version!!

Post by Frank Quisinsky » Wed Sep 27, 2017 4:55 am

Hi Lyudmil,

it's a rating list (KECR) for the first calculated move after the opening book ended. KECR calculated with 7.7 millions formulars.

Again: ONLY for the first move ... not more not less.

We try to make that with "how often the engine found the best move". Best move = the move most of 10 engines found.

So, not all of the 26.146 positions are good enough to build a "best move". Only very clear situations will go in the rating calculation.

--

And it is the development from a new opening book (FEOBOS). With other words, we try to make more as an easy opening book. The development of an Excel file for creating opening books for the future. We are thinking ... engines are stronger and stronger and for the books we can create we need a strong basic.

10 engines build a team itself for the opening book. Have a look in the readme files from the project (available on the FEOBOS site).

Best
Frank
I like computer chess!

Frank Quisinsky
Posts: 4852
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

Re: 506 rating list we produced with KECR

Post by Frank Quisinsky » Wed Sep 27, 2017 5:08 am

Hi Lyudmil,

For A00 - E99 ... to each of the codes an own Rating list.
Very important for ... which engine have the best results to all of the different ECO codes.

For A00 - A99, B00 - B99, C00 - C99, D00 - D99, E00 - E99
And for A00 - E99

The exactly answere must be:
= 506 x a rating list

And you can compare engine x with engine y and much more with KECR Excel file. I can say, no more questions to styles for the first playing Phase. I am looking in the Excel files and I found all what I need.

Please have a look in the Dia-Show to KECR, available on FEOBOS site.

Best
Frank
I like computer chess!

Frank Quisinsky
Posts: 4852
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

Re: FEOBOS v10 available ... first good version!!

Post by Frank Quisinsky » Wed Sep 27, 2017 5:32 am

Hi Norman,

I will write a German documentation to the end of the project. The documentation we have is the documantation for the two Excel files (KECR and FEOBOS) only.

All in German, Excel files are in German and English language. The English from Klaus and myself isn't good enough to write the docus in English language. But the website is in English and the Excel files are in English.

Cerebellum:
This is an opening book tuned with one engine only. In my opinion Stockfish is not the best engine for such a tuning because Stockfish calculated to often with draw. 15% from all (can be see in our project, grafic 9 in FEOBOS dia-show). No other engines in FEOBOS produced more 0,00 evals.

We try to create an opening book in teamwork by engines. Should be optimal for a rating list system. Balanced positions without bad moves and without positions, games can be ended with 3-fold very fast. Not longer as 3 moves after ECO codes formed. Cerebellum is much deeper as 3 moves only.

But with 3 moves after ECO codes formed and 10 engines and 41.000 positions one system need 1 year analyze time. With 4 moves after ECO codes formed one system need around 7 years analyze time with all the more positions.

Cerebellum is a great work but for one engine only, same I think with the Hiarcs books ... a great work for one engine only.

If you used such an opening book for other engines, other engines will be unlucky with around 10 - 15% of included lines. With FEOBOS and 10 engine analyzed the complete positions ... other engines will be unhappy with around 2-3%. Ca be see very easy on my start site. How many moves will be killed from each of the engines started for FEOBOS.

The main deal from FEOBOS:
An opening book for eng-eng matches, for a rating list system or engine tournaments ...

For engine - engine testing!!

FEOBOS isn't perfect for correspondence chess or online competitions. The book isn't deep enough for such things. Maybe only for checking ... is the position balanced 3 moves after ECO codes formed with the opinion by 10 engines ... can be see to each of the positions very easy in our FEOBOS Excel file.

In FEOBOS readme (available on FEOBOS site) I wrote ...

Code: Select all

Readme
 ###############################
Time of project:      13 months
 ~ Release date:    January 2018
###############################



A) BASICS: 26.146 positions ...

You can find here 26.146 positions of selected "FCP Rating List" games (~ 80%), newer missed
grandmasters theory games from 2014-2016 (~ 5%) and additional game material, most based on rare
opening lines (~ 15%).

Games are from selected "FCP Rating List" material based on my own "FCP Live Book", that
were under many years of development. With my own "FCP Live Book" engines try to find often
new lines on its own after the end of ECO code lines.

Summarized, my basics comes from:

- high quality TOP-50 computer chess games on Intel® Core™ i7-4770K hardware with 4Ghz and
  40 moves in 10 minutes with an Elo average from around ~ 2950 - 2975.
- high quality correspondence games
- high quality grandmaster games
  I am working most of times with Chessbase & TWIC databases.

 Before I start the project, my work files was cleaned and where analyzed with:

 1. Stockfish January 16th, 2017 BMI2 x64
   Setting: > 0.50 / -0.30
2. Komodo 10.3 x64
   Setting: > 0.50 / -0.30

- Intel® Core™ i7-4770K, 4.0 GHz
- 4 Cores / 8 Threads (Hyperthreading)
- Windows 7 Professional
- 4Gb for hash tables
 - 30 seconds per end position
- eval = 0.50 / -0.30


B) Upd.1: 15.468 positions ...

During the project, I saw different holes in my 26.146 database and I am searching in good
known theory relevant sources. I am building a new work file in combination with current
GM games on high level undo end of June 2017 and much other collected material from the
last months. Furthermore, more lines to rarely ECO codes searched / added.

Again, this "upd.1" work file was cleaned and where analyzed with the same engines I am
using for the 26.146 basic database:

1. Stockfish 8 BMI2 x64
   Setting: > 0.50 / -0.30
2. Komodo 10.3 x64
   Setting: > 0.50 / -0.30

- Intel® Core™ i7-4770K, 4.0 GHz
- 4 Cores / 8 Threads (Hyperthreading)
- Windows 7 Professional
- 4Gb for hash tables
 - 30 seconds per end position
- eval = 0.50 / -0.30

As result I can offer a database with 15.468 new positions (without move transpositions,
most 3 moves after ECO code was formed).

= 41.614 new basic positions without dublicates and without move transpositions.
= 93.794 new basic positions without dublicates and with move transpositions.



Main interest:
 --------------

- a strong PGN database for engine-engine matches / tourneys / rating list systems to
  balanced lines with clearly lesser probability in avoid fast draws (3-fold repetition).
- a strong PGN database for openings book developers.
  You can add the final FEOBOS result to your own opening book if you like.
- a strong PGN database for Arena Chess GUI "PGN-Random" feature.
 - a strong PGN database, positions are sorted with a ranking system (database without
  move transpositions) for building any Test-Set for engine research.

- opening book in Shredder Classic GUI *.bkt format
- opening book in Chessbase GUI *.ctg format
- opening book in Polyglot *.bin format

From time to time current versions of my opening books will be available during the
project is still running, called FEOBOS v1, v2 ...




Additional information:
-----------------------

- the database with move transpositions are included in project files
  (important for opening book vs. opening book testing).
- length of lines = most 3 moves after the ECO code was formed, minimal 4-6 moves.
  Please have a look in the file: "__eco-code-length.txt" (can be found in basics)
 - since FEOBOS v8 the opening books without move transpositions based on our own
  FEOBOS ranking system.
 - With analyses of "Experimental engine" (*.epd files) you can replace "Standard
  engines" if you like. With worksheet "Engines exp." you can compare results
  between "Standard engines" and "Experimental engines".




Used material:
--------------

- International opening keys by Sergiu Samarian, ISBN: 3-88805-299-8
- Small Encyclopedia of chess openings, ISBN: 978-86-7297-060-9
   http://www.chessinformant.org/small-encyclopedia-of-chess-openings-3rd-edition/
 - TWIC & different other sources of games!
  http://theweekinchess.com/

- Professional workbooks by Klaus Wlotzka
- 40H chess tools by Norm Pollock,  http://40h.000webhostapp.com/
  "epdOrder.exe", "epdList.exe", "epdInsert", "idOpcode.exe", "epdFin.exe", "pgnFin.cmd",
  "epdTrim.exe", "truncate.exe", "summary.exe" & "ecolist.exe", tagInsert.exe, txtZeros.exe
- PGN Extract by David J. Barnes,  https://www.cs.kent.ac.uk/people/staff/djb/pgn-extract/
- ELOstat by Dr. Frank Schubert,  http://www.amateurschach.de/main/_download.htm
- Tools by Ferdinand Mosca,  http://www.amateurschach.de/main/_download.htm
- Chessbase 14, http://www.chessbase.com
 - Fritz 15, http://www.chessbase.com
 - Shredder 13 by Stephan Meyer-Kahlen,  http://www.shredderchess.com
- Polyglot by Fabien LETOUZEY & Huang Chen & Harm Gert Muller &  Fonzy Bleumers
   http://hgm.nubati.net/polyglot.exe (Polyglot 2.0.3)
   https://github.com/ddugovic/polyglot
   http://hardy.uhasselt.be/Toga/polyglot-release/
   http://manpages.ubuntu.com/manpages/trusty/man6/polyglot.6.html
   https://chessprogramming.wikispaces.com/PolyGlot
  Polyglot Book Creator v0.1 by Ferdinand Mosca:
   http://www.talkchess.com/forum/viewtopic.php?topic_view=threads&p=711024&t=63635
  Polyglot Book Creator v0.2 by Ferdinand Mosca:
   http://www.talkchess.com/forum/viewtopic.php?topic_view=threads&p=711153&t=63635
- Notepad ++ by Dan HO,  https://notepad-plus-plus.org/

- OS: Windows 10 Professional, Windows 7 SP1 Professional x64
- Engines for analyses,  http://www.amateurschach.de/main/_engines.htm




 Helping persons:
----------------

- Klaus Wlotzka, Germany
  Developing of professional Excel workbooks!
- Ferdinand Mosca, Philippines
  "Game Extractor" to extract games in *.pgn format from a given pgn file!
  "Positions Explorer" and epd file with analysis or ce opcode!
  "SPF-Tool" to insert ranking number and sort *.pgn database to ranking number!
- Norm Pollock, USA 
  without Norm and his tools / helps many things here aren't possible!
 - Stefan Pohl, Germany
  some hints to the beginning of the project!

 Other helps and hints are welcome anytime!



August 24th, 2017 (22:30)
Frank Quisinsky, Trier (Germany)
I like computer chess!

Jeroen
Posts: 501
Joined: Wed Mar 08, 2006 8:49 pm

Re: FEOBOS v10 available ... first good version!!

Post by Jeroen » Wed Sep 27, 2017 7:19 am

In my view the idea that engines are able to conclude which opening positions are balanced and which ones are not, on the basis of a very short one-move-book-exit analysis, is simply wrong.

There are many openings in which engines indicate a (very) high score for one side, which turns out to be illusory after very deep analysis. These positions are actually very interesting to test for engine authors.

One other point: it is essential to include "complexity" in selecting good opening positions. And here the human factor is very important, as engines won't tell you which positions are 'complex' and which ones are 'easy'.

An example: a complex position with a +0.35 book exit is much more interesting to me, compared to a very easy position with a +0.20 book exit. The latter might have a 90+% probability of a draw, even between engines that have a 200 elo gap.

Post Reply