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.
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 01, 2018 8:39 pm

Hi Dann,

that are lines we produced in times we search for good positions (work bases). All the lines are not included in FEOBOS 41.614 basic database before 10 engines analyzed for bad moves.

Before we build the 41.614 FEOBOS positions, Stockfish and Komodo rejected bad lines. After this one 10 other engines with double hardware power analysed the complete 41.614 database. All analyses are available on FEOBOS Download Cockpit. Around 400 billions of nodes, more as one year of analysis. 500.000 analysis with 60 seconds by move from 10 different engines.

Interesting is FEOBOS v20.1 Contempt 3 database:
0 / 10 engines found a bad line and 2 / 10 engines max. give 0,00 eval.

= first 33.009 of 41.614 positions from our ranking system.

FEOBOS contempt 1-5 databases and books are online.

Best
Frank

10 engines found nothing in the first 33.009 positions with 4.3Ghz, 10 threads and 60 seconds per positions. So we are thinking the database is to 99,9% or higher free of errors.
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 01, 2018 9:06 pm

Hi Jeroen,

ABER MEIN BUCH IST DAS BESTE ...
I never said that to an own work in all the years computerchess. I am sure that you are waiting many years of such a comment by myself ... but sorry I can't help you.

For 18 years you wrote in Talkchess to an event by myself ...
Frank, you are a hero!

Jeroen, it’s important for our Hobby for many persons to have a bogeyman. Thinking on tournaments in Leiden ... you are the once persons in Leiden I never search a personally conversation. I have my reason for it!

German language is often not easy to understand.
For me the english language!

I wrote to Stefan in German language:

Code: Select all

Stefan Pohl sagt also:
Stockfish, Komodo, Houdini, Fire, Shredder, Wasp, Chrion, Texel, Andscacs und Booot ... alles nicht zu gebrauchen.
Aber mein Buch ist das Beste und mithin setzt sich Stefan Pohl über 3.000 Elo! 
That's my "Interpretation of Stefan's very aggressive Statements". Not more not less. If I read what Stefan wrote here, I can't build another interpretation. Í am the bogeyman for Stefan too, so you and Stefan can build a nice team.

But I am too old for such things!

If you are not able to understand what I wrote in German ...
It can be a clever idea not to set incorrect information in another forum.

Of course:
I accept your sorry!

:-)

Best
Frank

PS: I missed good arguments to FEOBOS in your sentence to our project. But sure, Stefan will like it to discuss with you about our project. I wish you and Stefan many fun.
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 01, 2018 9:57 pm

Hi Dann,

we can find very well bad moves with our system. It's right that if one of ten engines give a bad eval Position will be reject. We lost many good lines with it.

So, we developed a new idea for some weeks.

If one of ten engines only give a bad eval and eval isn't higher as 0.08 / -0.04 (we have a setting for it) as the by engine calculated automatically Setting, position go back to FEOBOS database. We called the idea "Rescue option".

If you are looking on start feobos worksheet you can see that 10 engines found 5.353 bad moves in 41.614 positions. With rescue 2.165 of such positions goes back in FEOBOS database.

We are working very tricky not to lose to many good positions for FEOBOS opening books.

Bad moves:
The worst 2.5% (we have a setting for it) of evals (75% white, 25% black) are bad moves. So, each of the ten engines found around 1.000 bad moves.

If you are looking in Start FEOBOS, the more engines analyzed the complete database the lesser is the result with bad moves. Because engines searching before found the most. That's very interesting because later all available engines should use FEOBOS books for rating systems.

Engine programmers know ... FEOBOS lines are analyzed by 10 engines and all is to 99,9% sure. If engines now lost very fast, engine programmer can better search and find mistakes because games are produced with a clean database.

So you can see that we need many engines for optimize opening books on a higher niveau. No human can do that because engines are much stronger as the World Champion. We say ... engines should do that work because humans are not strong enough for it, humans made to many errors.

Since engines are playing with 2.800 or more Elo, since this time opening experts in computer chess can go in Pension. Very hard but fact. No Person are able to create such strong books a group of 10 engines can do.

Thats my personal opinion.
A grandmaster Need a full day for a good opening analysis. We have 41.614 positions, a lot to do for a Grandmaster with around 600 elo lesser as the best engines with many cores today.

A lot of ideas we developed for FEOBOS and now we have a result. We can see which the best ECO Codes are, we can compare 41.614 Analysis of engines. Can directly see engines similars and so one.

Many additional information is produced and can be check in our Excel file.

But it's right ... sure that we lost good positions. Because, if two engines give a bad move the positions is no longer in our opening book. Probably not good enough for eng-eng matches.

The Advantage is:
A database free of bad moves and with a very high probably to avoid fast draw. That the deal we try as results. 10 engines in teamwork create a pure computer chess books.

All ideas we have to the Topic opening book are included in our Excel and the results is the ranking system with 8 criteria. Each Position have own own ranking. And on first positions in ranking are the positions if 10 enginse found many different moves. Because most interesting to see and for engines a good chance to tell us ... what can be the right way for such complicated positions.

Best
Frank
I like computer chess!

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

Re: Example: What we can do in the future of opening books!

Post by Frank Quisinsky » Thu Feb 01, 2018 10:37 pm

Hi Dann,

looking here!

You can see the automatic calculated engine setting.
If Fire in average are thinking 0.47 ... Houdini give 0.70 (for 2.5% of worst end positions in 41.614 FEOBOS database).

You can see FEOBOS TOP-50 in ranking and the evals by 10 engines, the average of 10 engine evals.

I think such information have to give a GUI as Output for modern opening books. Position 08.746 is on the board, user can directly see it's ranking position 2 in still running engine-engine match.

I think such information can build the future of opening book developments. If we have many computers and we can create such things much deeper as only three moves after ECO code was formed. We can create very strong criterias for a ranking system with strong developers.

With FEOBOS we build a new land ... think so!
Such information are much more interesting as stats to 1:0, 0:1 and draw to older GM games in opening books.

Best
Frank

Code: Select all

Eval setting max. =	0.70	0.50	0.47	0.62	0.76	0.58	0.74	0.68	0.57	0.56	
Eval setting min. =	-0.35	-0.31	-0.30	-0.39	-0.42	-0.28	-0.51	-0.36	-0.37	-0.32	

Ranking	Pos.-Nr.	Houd	Shred	Fire	Wasp	Chiron	Booot	Texel	Andsc	ASMFi	Komodo	avarage

1	Pos 21654	0.24	0.12	-0.08	0.13	0.11	0.06	0.25	0.09	-0.02	0.05	0.10
2	Pos 08746	0.38	0.15	0.21	0.44	0.35	0.14	0.18	0.47	0.02	0.28	0.26
3	Pos 37526	0.46	0.27	0.31	0.37	0.51	0.31	0.40	0.43	0.13	0.29	0.35
4	Pos 08011	0.06	0.13	0.13	0.17	0.26	0.10	0.07	0.12	0.01	0.20	0.13
5	Pos 26273	0.32	0.08	0.05	0.27	0.37	0.16	0.31	0.16	0.24	0.22	0.22
6	Pos 00313	0.43	0.11	0.16	0.07	0.32	0.05	0.40	0.34	0.24	0.16	0.23
7	Pos 05083	0.43	0.24	0.29	0.31	0.55	0.21	0.48	0.45	0.27	0.22	0.35
8	Pos 40559	0.53	0.39	0.37	0.47	0.70	0.43	0.69	0.59	0.31	0.39	0.49
9	Pos 09394	0.33	0.25	0.17	0.14	0.21	0.34	-0.16	0.29	0.29	0.15	0.20
10	Pos 10340	0.29	0.30	0.06	0.11	0.38	0.29	-0.03	0.30	0.43	0.28	0.24
11	Pos 33287	0.64	0.36	0.22	0.34	0.40	0.34	-0.24	0.32	0.31	0.29	0.30
12	Pos 08619	-0.12	-0.24	-0.13	-0.13	-0.20	-0.04	-0.05	-0.18	-0.07	-0.14	-0.13
13	Pos 31043	0.08	0.20	0.05	0.17	0.21	0.10	0.08	0.22	0.12	0.09	0.13
14	Pos 38143	0.16	0.09	0.11	0.17	0.07	0.22	0.17	0.28	0.07	0.20	0.15
15	Pos 25438	0.40	0.49	0.40	0.56	0.33	0.52	0.49	0.42	0.55	0.41	0.46
16	Pos 41310	0.23	0.11	0.13	0.11	0.33	0.13	0.24	0.36	0.17	0.16	0.20
17	Pos 11548	0.56	0.13	0.19	0.28	0.20	0.13	0.10	0.30	0.12	0.34	0.24
18	Pos 33622	0.37	0.13	0.17	0.28	0.22	0.26	0.31	0.36	0.20	0.22	0.25
19	Pos 32690	0.15	0.17	0.35	0.40	0.29	0.33	0.23	0.36	0.23	0.12	0.26
20	Pos 10678	0.19	0.31	0.28	0.41	0.43	0.26	0.25	0.40	0.06	0.12	0.27
21	Pos 31256	0.36	0.33	0.13	0.32	0.60	0.31	0.25	0.29	0.16	0.24	0.30
22	Pos 08235	0.15	0.35	0.16	0.46	0.50	0.35	0.16	0.36	0.34	0.20	0.30
23	Pos 31730	0.45	0.28	0.27	0.25	0.33	0.25	0.24	0.49	0.23	0.26	0.31
24	Pos 19446	0.46	0.18	0.22	0.21	0.26	0.32	0.25	0.50	0.29	0.39	0.31
25	Pos 10663	0.56	0.47	0.33	0.24	0.58	0.56	0.37	0.54	0.28	0.25	0.42
26	Pos 32406	0.31	0.07	0.05	0.24	0.33	0.15	-0.05	0.31	0.18	0.10	0.17
27	Pos 40028	0.43	0.31	0.18	0.31	0.58	0.28	0.25	0.59	0.36	0.39	0.37
28	Pos 21603	0.26	0.22	0.30	0.33	0.24	0.06	0.51	0.33	0.49	0.19	0.29
29	Pos 24007	0.26	0.08	0.16	0.31	0.24	0.23	0.31	0.31	-0.07	0.18	0.20
30	Pos 37533	0.08	0.16	0.27	0.33	0.57	0.33	0.37	0.55	0.08	0.21	0.30
31	Pos 25311	0.01	0.12	0.11	0.08	0.15	0.12	0.19	0.23	0.15	0.09	0.13
32	Pos 07605	0.13	0.04	0.03	0.10	0.34	0.24	0.07	0.18	0.08	0.11	0.13
33	Pos 02614	0.20	0.12	0.16	0.13	0.16	0.25	0.03	0.17	0.22	0.22	0.17
34	Pos 05977	0.10	0.26	0.29	0.26	0.44	0.03	0.09	0.20	0.15	0.06	0.19
35	Pos 35164	0.21	0.13	0.13	0.18	0.20	0.29	0.13	0.34	0.25	0.16	0.20
36	Pos 29529	0.24	0.04	0.07	0.22	0.31	0.25	0.48	0.33	0.08	0.23	0.23
37	Pos 28534	0.21	0.27	0.19	0.17	0.47	0.26	0.30	0.24	0.22	0.28	0.26
38	Pos 05048	0.32	0.31	0.11	0.44	0.47	0.27	0.39	0.31	0.30	0.17	0.31
39	Pos 41215	0.33	0.47	0.36	0.23	0.37	0.32	0.26	0.33	0.33	0.13	0.31
40	Pos 06241	0.35	0.40	0.29	0.29	0.51	0.31	0.18	0.28	0.33	0.20	0.31
41	Pos 21493	0.38	0.32	0.13	0.18	0.57	0.32	0.47	0.49	0.16	0.41	0.34
42	Pos 06075	0.32	0.34	0.42	0.36	0.65	0.47	0.39	0.39	0.38	0.25	0.40
43	Pos 27970	0.40	0.30	0.36	0.35	0.73	0.39	0.48	0.46	0.27	0.44	0.42
44	Pos 04919	0.31	0.26	0.35	0.50	0.56	0.46	0.69	0.47	0.24	0.53	0.44
45	Pos 13709	0.56	0.23	0.33	0.51	0.63	0.45	0.50	0.59	0.28	0.35	0.44
46	Pos 05366	0.51	0.44	0.27	0.47	0.65	0.37	0.31	0.53	0.39	0.50	0.44
47	Pos 03947	0.65	0.42	0.32	0.42	0.61	0.48	0.40	0.50	0.47	0.51	0.48
48	Pos 06867	0.56	0.43	0.37	0.61	0.74	0.43	0.67	0.61	0.57	0.46	0.55
49	Pos 27565	0.08	0.12	0.20	0.16	0.18	0.18	0.16	0.19	0.11	-0.01	0.14
50	Pos 19197	0.18	0.15	0.15	0.15	0.35	0.21	0.20	0.23	0.24	0.21	0.21
Such Things can be generate with our Excel.
Positions 1 ... 10 engines found 9 different best moves (E11 position). So you can generate each thinkable statistic with all the analysis.

And now looking in Detail ...
No 0,00 evals ... because we are speaking from the TOP-50 ranking positions. If one engine give 0,00 ... Contempt 4, if two engines comes with 0,00 ... Contempt 3.

Very complicated to explain all the ideas we have in English. I working on a documentation in German language. Site 26 is written and I am sure over 50 sites if I am ready, so many things we do in the last year. Many material for the documentation.
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 01, 2018 11:32 pm

Looking here:
Last 1000 of 41.614 positions based on a mistake by myself and are rejected (lines inside an ECO code with unequal move length ... a long toptic, for the Moment not interesting). But looking before the 1.000 lines on the end of our ranking system ...

The bad lines ...
Such lines I will not see for eng-eng matches. The advantage is to high. Weaker engines can make draw vs. clearly stronger engines and statistic results will be unclear. For eng-eng an absolutely strong and checked database is necessary.

All is good sorted with our ranking system!

Best
Frank

Code: Select all

40555	Pos 25688	0.36	0.62	0.45	0.70	1.02	0.62	1.17	0.74	0.35	0.45	0.65
40556	Pos 14134	-0.31	-0.37	-0.36	-0.70	-0.50	-0.27	-0.21	-0.43	-0.42	-0.30	-0.39
40557	Pos 31462	0.76	0.53	0.42	0.63	0.72	0.60	0.63	0.73	0.57	0.68	0.63
40558	Pos 31574	0.69	0.64	0.60	0.71	0.93	0.44	0.53	0.74	0.37	0.66	0.63
40559	Pos 27125	-0.45	-0.20	-0.22	-0.40	-0.48	-0.28	-0.34	-0.41	-0.38	-0.37	-0.35
40560	Pos 39777	-0.36	-0.28	-0.25	-0.40	-0.40	-0.37	-0.51	-0.48	-0.45	-0.36	-0.39
40561	Pos 36469	-0.48	-0.30	-0.16	-0.41	-0.54	-0.32	-0.36	-0.38	-0.27	-0.44	-0.37
40562	Pos 38789	-0.31	-0.31	-0.40	-0.41	-0.48	-0.39	-0.54	-0.61	-0.20	-0.27	-0.39
40563	Pos 35601	0.55	0.57	0.56	0.52	0.78	0.51	0.86	0.64	0.72	0.59	0.63
40564	Pos 11486	0.61	0.72	0.46	0.92	0.81	0.85	0.60	0.98	0.62	0.29	0.69
40565	Pos 38494	0.39	0.42	0.53	0.70	0.88	0.71	0.80	0.66	0.27	0.79	0.62
40566	Pos 41317	0.55	0.53	0.68	0.85	0.66	0.49	1.11	0.75	0.62	0.39	0.66
40567	Pos 13084	0.81	0.62	0.57	0.64	0.69	0.67	1.03	0.66	0.51	0.53	0.67
40568	Pos 12248	-0.50	-0.28	-0.47	-0.51	-0.42	-0.36	-0.49	-0.48	-0.37	-0.35	-0.42
40569	Pos 38162	0.35	0.51	0.52	0.53	1.07	0.65	1.12	0.72	0.51	0.31	0.63
40570	Pos 13053	0.58	0.58	0.53	0.64	0.82	0.73	0.54	1.03	0.52	0.44	0.64
40571	Pos 18910	-0.18	-0.30	-0.22	-0.41	-0.53	-0.17	-0.52	-0.39	-0.42	-0.33	-0.35
40572	Pos 13636	-0.41	-0.75	-0.47	-0.73	-0.38	0.00	-1.02	-0.18	-0.40	-0.12	-0.45
40573	Pos 35687	-0.28	-0.35	-0.37	-0.41	-0.45	-0.45	0.00	-0.52	-0.17	-0.24	-0.32
40574	Pos 32943	-0.28	-0.38	-0.47	-0.49	-0.56	-0.22	-0.66	-0.51	0.00	-0.19	-0.38
40575	Pos 00807	0.03	-0.45	-0.37	-0.58	-0.67	0.00	-0.62	-0.39	-0.28	0.00	-0.33
40576	Pos 13579	-0.03	-0.42	-0.64	-0.91	-0.44	0.00	-1.46	-0.54	0.00	0.17	-0.43
40577	Pos 34747	-0.44	-0.87	0.00	-0.48	-0.41	0.00	-0.74	-0.43	0.00	-0.37	-0.37
40578	Pos 30318	0.82	0.52	0.50	0.65	0.72	0.69	0.97	0.68	0.47	0.63	0.67
40579	Pos 04114	0.66	0.72	0.56	0.78	1.00	0.44	1.14	0.85	0.54	0.74	0.74
40580	Pos 41346	0.81	0.67	0.55	0.50	0.92	0.52	0.95	0.69	0.63	0.26	0.65
40581	Pos 41483	0.67	0.48	0.56	0.74	0.77	0.47	0.85	0.71	0.65	0.61	0.65
40582	Pos 25579	0.74	0.51	0.54	0.65	0.72	0.58	0.84	0.86	0.54	0.60	0.66
40583	Pos 29874	0.34	0.59	0.55	0.78	1.06	0.34	0.67	0.69	0.66	0.66	0.63
40584	Pos 29335	0.87	0.59	0.42	0.56	0.83	0.79	1.12	0.97	0.96	0.51	0.76
40585	Pos 13651	0.71	0.69	0.47	0.68	0.99	0.66	0.01	0.86	0.59	0.55	0.62
40586	Pos 40659	0.77	0.56	0.29	0.88	0.88	0.61	0.99	0.51	0.42	0.57	0.65
40587	Pos 29003	0.65	0.66	0.51	0.82	0.94	0.62	0.78	0.83	0.44	0.50	0.68
40588	Pos 13637	-0.33	-0.60	-0.40	-0.52	-0.57	-0.20	-0.96	-0.40	-0.41	-0.14	-0.45
40589	Pos 41176	0.78	0.51	0.37	0.68	0.78	0.59	0.51	0.68	0.60	0.73	0.62
40590	Pos 34916	0.73	0.61	0.47	0.21	0.48	0.64	0.78	0.69	0.84	0.61	0.61
40591	Pos 26338	-0.24	-0.26	-0.38	-0.58	-0.92	-0.34	-0.66	-0.46	-0.32	-0.34	-0.45
40592	Pos 36823	0.73	0.48	0.56	0.69	0.67	0.66	0.32	0.77	0.62	0.63	0.61
40593	Pos 04443	0.79	0.54	0.44	0.55	0.83	0.61	0.80	0.79	0.43	0.64	0.64
40594	Pos 41429	0.79	0.57	0.58	0.71	0.73	0.77	1.00	0.98	0.39	0.44	0.70
40595	Pos 40095	0.71	0.64	0.22	0.67	0.41	0.28	0.79	0.76	0.77	0.68	0.59
40596	Pos 26112	0.53	0.60	0.73	0.67	0.79	0.72	0.84	0.59	0.30	0.71	0.65
40597	Pos 30282	0.43	0.57	0.52	0.72	0.82	0.68	0.91	0.80	0.42	0.55	0.64
40598	Pos 31756	0.82	1.17	0.68	0.76	0.96	0.45	1.07	0.33	0.71	0.00	0.70
40599	Pos 31237	0.90	1.02	0.83	0.83	0.00	0.66	0.00	1.06	0.74	0.34	0.64
40600	Pos 21067	-0.60	-0.26	-0.43	-0.51	-0.47	-0.44	-0.30	-0.48	-0.46	-0.42	-0.44
40601	Pos 41381	0.68	0.58	0.63	0.87	0.80	0.68	0.81	0.81	0.49	0.57	0.69
40602	Pos 29691	0.70	0.51	0.54	0.68	0.85	0.60	0.76	0.71	0.49	0.58	0.64
40603	Pos 28940	0.89	0.80	0.70	1.22	1.05	0.75	1.68	0.94	0.52	0.51	0.91
40604	Pos 28973	1.02	0.53	0.60	0.55	0.71	0.70	0.82	0.79	0.63	0.81	0.72
40605	Pos 23296	0.93	0.50	0.70	0.70	0.78	0.60	0.84	0.58	0.59	0.57	0.68
40606	Pos 34360	0.77	0.55	0.48	0.69	0.94	0.73	0.67	0.69	0.56	0.57	0.67
40607	Pos 31738	-0.36	-0.28	-0.32	-0.47	-0.53	-0.35	-0.59	-0.58	-0.45	-0.29	-0.42
40608	Pos 34004	0.81	0.66	0.79	0.67	0.84	0.86	0.85	0.91	0.54	0.51	0.74
40609	Pos 29243	0.87	0.54	0.38	0.66	0.77	0.59	0.61	0.81	0.89	0.81	0.69
40610	Pos 09263	0.93	0.56	0.70	0.34	0.90	0.59	1.13	0.64	0.99	0.72	0.75
40611	Pos 04372	0.77	0.51	0.55	0.63	0.88	0.72	0.84	0.76	0.46	0.66	0.68
40612	Pos 04380	0.78	0.55	0.50	0.54	0.98	0.78	0.88	0.89	0.62	0.72	0.72
40613	Pos 41614	0.71	0.49	0.78	0.98	0.90	0.66	0.91	1.18	0.80	0.87	0.83
40614	Pos 13638	-0.39	-0.50	-0.37	-0.71	-0.49	-0.27	-0.85	-0.44	-0.42	-0.35	-0.48
I like computer chess!

Dann Corbit
Posts: 10756
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 4:32 am

Frank Quisinsky wrote:Hi Dann,

that are lines we produced in times we search for good positions (work bases). All the lines are not included in FEOBOS 41.614 basic database before 10 engines analyzed for bad moves.

Before we build the 41.614 FEOBOS positions, Stockfish and Komodo rejected bad lines. After this one 10 other engines with double hardware power analysed the complete 41.614 database. All analyses are available on FEOBOS Download Cockpit. Around 400 billions of nodes, more as one year of analysis. 500.000 analysis with 60 seconds by move from 10 different engines.

Interesting is FEOBOS v20.1 Contempt 3 database:
0 / 10 engines found a bad line and 2 / 10 engines max. give 0,00 eval.

= first 33.009 of 41.614 positions from our ranking system.

FEOBOS contempt 1-5 databases and books are online.

Best
Frank

10 engines found nothing in the first 33.009 positions with 4.3Ghz, 10 threads and 60 seconds per positions. So we are thinking the database is to 99,9% or higher free of errors.
The exit positions are all pretty safe. But there are positions inside of the book before you exit that are up or down a full piece (and not just extreme gambits).

So whether this is a problem or not depends on what you use the book for.

If you use it to play engines which are not using the same book, then you will get some of the extreme positions actually played and then it will matter.
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.

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 » Fri Feb 02, 2018 7:07 am

Hi Dann,

I understand what you mean.

First:
FEOBOS is for using eng-eng match constallation, rating list systems for an example. All engines have to start with the same opening book and easy prioritys with balanced positions the games.

Second:
At first I am thinking I have to simulate PGN random option from Arena. No other GUI used my idea with PGN random (by random a position from a PGN file is loaded up to move x, for eng-eng games. Arena saved the source of the eng-eng games the new *.pgn file. In the beginning of Arena I had the opinion, book options are for eng-eng not necessary. PGN random is much more interesting if a strong *.pgn database will be available. Thats the main reason I try with Shredder GUI and the possibilites I have here to simulate PGN random with book options by Shredder GUI.

Third, to your question:
We have a lot of such books you have written before. I have created a lot of such books too, Gambit-Books for an example, books you have the right answer of bad moves and so one. Interesting to have it if engines will used an own opening book.

What we not have is such a strong database engines created automacially with FEOBOS. I am working since a longer time with Shredder GUI. With Shredder very easy to solved.

Create a new opening book with Shredder GUI. Add the FEOBOS lines. Convert the work in Shredder's text format. Set by an editor all priorities to A. Convert the book back to Shredder's *.bkt format. All FEOBOS lines have for the moment priority A. Now I added all material you need from not balanced moves in the the same book. Convert to *.txt and set for lines with non priorities the priority E (E will not play by Shredder for the moment). No convert back to Shredder's *.bkt format.

Now you have a book, all balanced lines have prio "A" and all not balanced lines have prio "E". Now you have by hand to set for ...

Prio "A" ... more prios "A-D, popular or not popular, can checked missed lines very easy because all what is missed will have prio "E".

Thats most work for professional opening books, should be have for each bad move the right answer.

Same you can do with killer lines, for killer lines you can give at first priority F for an example.

But this isn't topic for FEOBOS and in my opinion very boring. A very easy work most of book creators do with all the possible tricks we have.

To have a error free database is much more interesting for eng-eng match constellation because we don't have a *.pgn file without any errors ... or better checked by 10 engines.

In my opinion very important to have it because with a Elo calculation system we can create much better ratings of engines. The main deal of the work I do with Klaus in development FEOBOS.

Best
Frank

In FEOBOS we have many options. One of the options is that FEOBOS can give each ECO code a priority. But again, we have enough of such books, we missed a strong and clear, error free database for eng-eng matches. I have more interest to generate this one or better I have interest that engines generate this one in team work!
I like computer chess!

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

Re: Eldorado for any test-set constellation ...

Post by Frank Quisinsky » Fri Feb 02, 2018 7:21 am

Hi Dann,

with a by ranking sorted database with balanced material to all ECO codes you can generate any test-set constellation !!

Give me the best B01, B02, C74 positions and so one. First positions to B01 in database is best position to B01 with FEOBOS ranking system.

So you can do this one:
Give me the best 20 B01 and you can test with eng-eng and our balanced FEOBOS database which engine is playing B01 with most success.

OK can be see for all ECO Codes in our second Excel ... KECR too. Which engines understand all the different ECO codes more as the others.

FEOBOS is more or less a database with 33.009 test positions. You can a lot do with such a database. Using for test-set constellation or create an opening book, using for PGN Random for Arena GUI and so one. You can added FEOBOS to developed opening books and such books for closing all the holes to balanced lines.

Best
Frank
I like computer chess!

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

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

Post by Jeroen » Fri Feb 02, 2018 3:24 pm

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?

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

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

Post by Jeroen » Fri Feb 02, 2018 4:28 pm

As the c) lines contain moves that are weak, these lines are not interesting. A weaker move by one side is cancelled out by a weaker move by the other side, making it 'playable' according to the engines evals. Humans will never select such a line, though.

Maybe the b) lines are of some interest to people who want to avoid theory and whom like to know if (some of) these lines are playable. But you'll need more lines and deeper analysis to be sure, not very shallow searches. A deeper analysis might reveil that the lines might not be that good.

Regarding the a) lines: thousands of people have already spent (tens of) thousands of analysis hours on them, using iDEA, automatic book development, testing and so forth. They are nothing special. In an engine-engine match these lines simply will be followed up by more theory moves.

If I had the choice between a FEOBOS shallow search by 10 engines and a very deep 2-5 months iDEA analysis on a position, of course I would select the latter. 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.

Post Reply