Release: FEOBOS v20 Contempt 1-5 without move transp. avail.

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.
User avatar
Ovyron
Posts: 3963
Joined: Tue Jul 03, 2007 2:30 am

Re: Release: FEOBOS v20.1 available, sorry!

Post by Ovyron » Fri Feb 02, 2018 7:09 pm

Jeroen wrote:And if my project would only allow 15 minutes on each position, I would pick SF, Houdini and Komodo and let them calculate 5 mins on each position.
I'd do 7.5 mins split between Houdini and Stockfish.

Unfortunetely, most of the positions will be full of redundancy where you just wasted 10 minutes checking what you already knew, what matters are the critical positions that refute an entire line, and for those the extra 2.5 minutes for each engine might get critical if they all would have missed the move in 5 minutes.

But yeah, using 10 engines at 1.5 mins per move would seem like a great waste of resources.

Dann Corbit
Posts: 10781
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Release: FEOBOS v20.1 available, sorry!

Post by Dann Corbit » Fri Feb 02, 2018 7:40 pm

Jeroen wrote:
I see another big problem: The spit out evals are not stable.
Yes. A +0.20 might turn into a +0.50 if you double, triple or quadruple the time. And a -0.50 might only be -0.15 when given more time.

BTW, what do you do if the 3 strongest engines say +0.20 and the others +0.50? Skip the line? Or follow the opinion of the stronger programs?
I have just started to analyze the entire pile to depth (both interior and terminal nodes).

I suggest that Stockfish-9 to 36 plies should be pretty good evidence of the value of the position.

Of course, I do this for every new important book that comes out.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Dann Corbit
Posts: 10781
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Release: FEOBOS v20.1 available, sorry!

Post by Dann Corbit » Fri Feb 02, 2018 7:42 pm

Ovyron wrote:
Jeroen wrote:And if my project would only allow 15 minutes on each position, I would pick SF, Houdini and Komodo and let them calculate 5 mins on each position.
I'd do 7.5 mins split between Houdini and Stockfish.

Unfortunetely, most of the positions will be full of redundancy where you just wasted 10 minutes checking what you already knew, what matters are the critical positions that refute an entire line, and for those the extra 2.5 minutes for each engine might get critical if they all would have missed the move in 5 minutes.

But yeah, using 10 engines at 1.5 mins per move would seem like a great waste of resources.
Unless you are one of the engine authors. In that case, it is an incredible treasure trove of information.

And, on some occasions, a weaker engine will see something that the stronger one missed.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

tpoppins
Posts: 919
Joined: Tue Nov 24, 2015 8:11 pm
Location: upstate

Re: Release: FEOBOS v20.1 available, sorry!

Post by tpoppins » Tue Feb 06, 2018 11:27 pm

First, I'd like to state that I agree with the argument that engine analysis of openings, taken on its own, is largely meaningless.

However, I don't think that broad, sweeping statements like "nobody needs FEOBOS" are justified. Dismissing someone's hard work without considering the whole picture is somewhat disrespectful and speaking on everybody's behalf sounds arrogant and presumptuous.

I am one of those nobodies that need FEOBOS. I use other books and test suites as well, but this one is my favorite by far. The main reasons for that are:

1) FEOBOS offers variety like few other books. There are even larger books -- GM 8-move (I forget the exact filename) is probably the biggest -- but they are full of lines where one side emerges with a 100+ cp advantage or that are liable to end in a draw five moves out of the book when played by top engines. Other highly-acclaimed books/test sets would have you believe there's little more to opening theory besides a handful of lines in Giuoco Piano, Caro-Kann, Ruy Lopez, Nimzo-Indian, Semi-Slav and QGD.

Not so with FEOBOS -- thousands of well-balanced lines with plenty of scope for creative play. Some people are happy to use the same 50-line test suite year after year. Some people are also OK with having a blank wall as the view out of their window. Others prefer to see the expanse of fields receding towards the woods/mountains, or perhaps the sea. If limitless horizons are your cup of tea, FEOBOS is the book for you.

2) Surprisingly low draw rate considering the average length of lines. I used FEOBOS in more than 30 tests for CCRL and here are the stats for the three books I used most:

Code: Select all

book           |  number of short draws | total games | short draws ratio
-------------------------------------------------------------------------
HERT500        |          15            |    3,170    |      0.47%
Noomen 3-move  |          49            |    8,670    |      0.57%
FEOBOS 		  |          97            |   27,622    |      0.35%
with short draws meaning games drawn in 20 moves or less. I'm posting this not in order to claim that FEOBOS is better than, say, Noomen 3-move, just to show that the former manages to improve on the latter's already very low draw rate by a factor of 1.6.

I'd like to use the new SALC books as well; however, the larger of the two consists entirely of lines that at 10 moves are a bit too long for rating lists. The general preference these days seems to be for much shorter books, like Noomen's 2-moves and 3-moves. I'll probably start using the smaller closed-positions-only test suite; although quite small at 500 lines, the premise appears very interesting.

Big thanks to Frank and Co. for the great book! You've done a terrific job and I hope you will consider releasing a shorter version just for rating lists.

The version I've been using for the past couple of months is feobos_v10-contempt-4.pgn edited down to 12 plies (with the resulting duplicate lines removed with PGN Tools). Now that Dann listed some lines where balanced book exits are reached via some extremely unbalanced positions I'm a little worried that some of the book exits in my edited version may have come unhinged that way. I haven't seen any lopsided results where White (or Black) would get short wins in both games of a round played with the same opening with colors reversed, but it's a potential concern.

Dann Corbit
Posts: 10781
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Release: FEOBOS v20.1 available, sorry!

Post by Dann Corbit » Wed Feb 07, 2018 11:59 pm

I will give updates on my analysis for the interior and exterior book nodes.

For example:
http://rybkaforum.net/cgi-bin/rybkaforu ... pid=577815
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Dann Corbit
Posts: 10781
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Example: The bad lines ...

Post by Dann Corbit » Thu Feb 08, 2018 9:09 am

I like what you are doing and I find it interesting.
It also gives me new ideas to explore.
I found a few things that make me smile.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

User avatar
Kotlov
Posts: 216
Joined: Fri Jul 10, 2015 7:23 pm
Location: Russia

Re: Release: FEOBOS v20.1 available, sorry!

Post by Kotlov » Thu Feb 08, 2018 1:42 pm

Good work!

Frank Quisinsky
Posts: 4875
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

Re: Release: FEOBOS v20.1 available, sorry!

Post by Frank Quisinsky » Thu Feb 08, 2018 7:25 pm

Hi,

10 Engines x 41.614 Positions x 60 Seconds = 299 Billions Nodes.
Intel 10 Threads, 6 Cores with 4.3 GHz.

More isn't possible for me! Double time = 2 years analyze time!!

Best
Frank
I like computer chess!

Frank Quisinsky
Posts: 4875
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

Re: Release: FEOBOS v20.1 available, sorry!

Post by Frank Quisinsky » Thu Feb 08, 2018 7:51 pm

Hi,

thanks for your interest.
Yes, you understand the idea of the FEOBOS project.

3 years I try to collect positions and worked hard on FCP Live Book. The time the FCP Rating List was still running. A lot of material comes from TOP-56 engines. But material from TOP Players are included too, from correspondence chess and ... the balanced positions from "Small Encyclopaedia of chess openings ... best material we have to openings in my opinion". Most of material from Upd.1 is to rarely opening keys and missed material comes from Enclyoapedia of Chess openings. So I can fill the holes with upd.1.

A wild mix what I collect for FEOBOS. Most problem is that for many ECO codes not many balanced material is possible.

Shortly:

You have a lot to games undo move 20. It would be great to have this games. I have a database with such games too.

Comes from:
3-moves after ECO Code formed and many ECO Codes formed very fast. We have lines ended with 5 moves in FEOBOS. Undo move 20 is a long way and most short draw lines comes from this ECO codes.

Most ECO Codes with 8-12 moves ... here you can see that games ended with draw undo move 20 are very rarely.

So the team work of 10 engines can produced here TOP statistics with our system to reject "draw lines".

I am thinking:
Komodo vs. 300 Elo lesser ...
If the Opponent from Komodo have a small Advantage with the start position engine can Play more draws vs. Komodo and finally the stastitic result is bad because weaker engine can make more points.

With FEOBOS better statistics can be produced for rating systems.

Shorter Version:
FEOBOS database is sorted by ranking !!
Best of 41.614 on first.
Optimal is Contempt 3 because Booot and Stockfish give to many 0,00 evals and with Contempt 5 to many good positions lost.

First 33.009 of 41.614 positions are good for engine testing. You can go higher ... Contempt 4 or Contempt 5 if you like a smaller database of positions.

Please not edit to 12 plies. Many ECO Codes formed after 14 moves and FEOBOS is 3 moves after ECO Codes was formed. So you lost many ECO codes for testing if you used only 12 plies.

Additional Information:
You can see in TXT files, available in download files ... also sorted with ranking (same as PGN file) all the evals by engines. The first 33.009 from the final database are to 100% free of bad moves !!

For two days I added opening book 6 ...
The book with move transpositions and Contempt 3 (without rescue lines). So you can files holes the book have with all the move transpositions to the way to the end position.

Ferdinand programmed a tool for us to find all draw lines and bad lines in the bigger databases with move Transposition.

Please donwload Version 20.1 again and you have all 6 opening books.

Best
Frank

Again, thanks for your nice words.
We developed a lot of new ideas in Excel and the Event is the ranking Systems. The FEOBOS databases is sorted by ranking, each position have an own ranking. Thats the Highlight for the final Version 20.1.
I like computer chess!

Frank Quisinsky
Posts: 4875
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

Re: Example: The bad lines ...

Post by Frank Quisinsky » Thu Feb 08, 2018 8:00 pm

Hi Dann,

most interesting is the FEOBOS ranking System. All positions have an own ranking after 8 criterias. And *.pgn database is sorted by ranking.

First 33.009 of FEOBOS database are the best for testing engines (Contempt 3 database).

Sine two days FEOBOS v20.1 opening book 6 with move transpositions is available. Ferdinand programmed again a tool for us. So I can delete draw / bad moves in the database with move transpositions.

Please download again FEOBOS v20.1 ... 6 books now and database with move transpositions is available.

The database with move transpositions fill the holes you have spoken before a lot.

:-)

Best
Frank
I like computer chess!

Post Reply