Hi there,
each of the 26.146 balanced positions (three moves after ECO code formed) have now an own ranking in FEOBOS Excel files.
Easy to find out best positions to each of the ECO codes for testing engines.
Now FEOBOS 5.06 Test-Set is ready and checked with a Komodo - Stockfish match (produced 68,2% draws only). So FEOBOS works great because positions with a danger for a fast draw rejected in FEOBOS.
For each of the 500 ECO code 1 position = 1.000 games (white / black).
Can be found on FEOBOS site under:
http://www.amateurschach.de/main/_new-opening-book.htm
Best
Frank
PS: Today we aren' strong enough for creating opening books for testing engines. Engines are 500-1800 Elo stronger. Let the engines do that work itself = FEOBOS!
FEOBOS 5.06 Test-Set (A00-E99) available ...
Moderators: hgm, Rebel, chrisw
-
- Posts: 6808
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
-
- Posts: 1080
- Joined: Fri Sep 16, 2016 6:55 pm
- Location: USA/Minnesota
- Full name: Leo Anger
Re: FEOBOS 5.06 Test-Set (A00-E99) available ...
Your project sounds very interesting and fun.
Advanced Micro Devices fan.
-
- Posts: 6808
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: FEOBOS 5.06 Test-Set (A00-E99) available ...
Hi Leo,
oh yes ... we have many fun here with the Excel workbook dev. and all the analysis by engines.
Isn't the final version.
End of the project I can create the final version.
We found a problem in our ranking (each position have an own ranking in Excel workbook).
Criterium 1: No bad moves
Criterium 2: No draw moves
Criterium 3: Queens on board (should be included ... 384 of 26.146 positions are without queens on board ... not very interesting for a test-set. In the test-set I created all positions with queens on board.
Criterium 4: Most different first moves (at first criterium 1 before, that was wrong)
Criterium 5: Have to do with experimental engines
Criterium 6: average of eval of analyzed engines
So we need more engine analysis for a better ranking.
In Excel file you can sort ... sort by ECO and Ranking and you can see directly best, second best, third best ... positions for each of the codes.
The test-set I created at first based on an older ranking system, not perfect.
With other words:
The test-set is very good but can be better with a better ranking system and more engine analysis.
Look here:
http://www.amateurschach.de/ftptrigger/ ... t%202.html
Differnet after 2.000 games vs. 20 opp. = 20 Elo (20 Elo better the newer Wasp dev. version).
After the current version with my test-set and around 360 games = 26 Elo.
End of the project I will do it again ... think so!
Best
Frank
A lot of work for Klaus to changed that at the moment.
But if the ranking works fine ...
Very easy ... To the end of the project you can say ...
1. Please the first 22.000 / 26.146 positions (after ranking calculation) for the opening book.
2. Please the best 100, 200, 500, 1000 positions after ranking for a test-set.
3. Please the best position for each eco Code for a test-set (what I do).
4. You are a B01 fan ... so you can say ... please the best 20 B01 for a B01 test-set ...
5. ...
6. ...
oh yes ... we have many fun here with the Excel workbook dev. and all the analysis by engines.
Isn't the final version.
End of the project I can create the final version.
We found a problem in our ranking (each position have an own ranking in Excel workbook).
Criterium 1: No bad moves
Criterium 2: No draw moves
Criterium 3: Queens on board (should be included ... 384 of 26.146 positions are without queens on board ... not very interesting for a test-set. In the test-set I created all positions with queens on board.
Criterium 4: Most different first moves (at first criterium 1 before, that was wrong)
Criterium 5: Have to do with experimental engines
Criterium 6: average of eval of analyzed engines
So we need more engine analysis for a better ranking.
In Excel file you can sort ... sort by ECO and Ranking and you can see directly best, second best, third best ... positions for each of the codes.
The test-set I created at first based on an older ranking system, not perfect.
With other words:
The test-set is very good but can be better with a better ranking system and more engine analysis.
Look here:
http://www.amateurschach.de/ftptrigger/ ... t%202.html
Differnet after 2.000 games vs. 20 opp. = 20 Elo (20 Elo better the newer Wasp dev. version).
After the current version with my test-set and around 360 games = 26 Elo.
End of the project I will do it again ... think so!
Best
Frank
A lot of work for Klaus to changed that at the moment.
But if the ranking works fine ...
Very easy ... To the end of the project you can say ...
1. Please the first 22.000 / 26.146 positions (after ranking calculation) for the opening book.
2. Please the best 100, 200, 500, 1000 positions after ranking for a test-set.
3. Please the best position for each eco Code for a test-set (what I do).
4. You are a B01 fan ... so you can say ... please the best 20 B01 for a B01 test-set ...
5. ...
6. ...