Deed's Chess Learning System (Moderation actions)

[deeds]

Thanks to DeepSeek AI (38%) !

[deeds]

...
As these engine releases are no longer compatible with the new sfnnv6 nets i will avoid publishing the results of my next learning sessions in a general public area.
Out of respect for the enlightened and customers, i will publish them on the webshop (see : computerchess => learning sessions).
...

LS14 completed !

The distribution of trained openings is rebalanced after each learning session :

[deeds]

Openings to learn
For my 15th learning session, i use a learning farm with 4 trainings at the time :

opening #1 = (C65) 1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. d3 Bc5 5. Bxc6 dxc6 6. O-O Nd7 7. Nbd2 O-O 8. Nc4 Re8 9. Bd2 f6 10. Nh4 Bf8 *
opening #2 = (C65) 1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. d3 d6 *
opening #3 = (E05) 1. d4 Nf6 2. c4 e6 3. g3 d5 4. Bg2 dxc4 5. Nf3 Be7 6. O-O O-O 7. Qc2 a6 8. Qxc4 b5 9. Qc2 Bb7 10. Bf4 Nc6 11. Rd1 *
opening #4 = (E41) 1. Nf3 Nf6 2. c4 e6 3. Nc3 Bb4 4. e3 c5 5. d4 Bxc3+ *


Training with books
Eman uses +70 books as sparring-partners :



Reinforcing experience data
Eman analyses at D40 some interesting games, played with sparring-partners like SF15.1, SF16.1 or books, which unexpectedly ended in a draw :



Bruteforcing books
Eman analyses all the legal moves allowed by the book's lines :

[deeds]

Creating EXP files
Previously, I created an EXP file from Rybka's games.
Then, i ran 800 games with Rybka against its EXP file.
Now Rybka has to replay all the openings it lost :



Testing EXP files
I use Eman 8.40 to improve/check the accuracy of my Depth4 EXP file with stronger engine like SF 15.1, SF16.1 or TC odds.

[deeds]

Bruteforcing books
Eman analyses all the legal moves allowed by the book's lines :

Bruteforce attack on ROS 2024.bin is over !
+1,99/26 @ 21%

[deeds]

Creating EXP files
Previously, I created an EXP file from Rybka's games.
Then, i ran 800 games with Rybka against its EXP file.
Now Rybka has to replay all the openings it lost :

The creation of the Rybka's EXP file is over !
from -8 to +49 elo

[deeds]

Results of the experiment :



Modern engines (NNUE) require very precise experience data to obtain a gain from an EXP file.
Older engines (without NNUE) obtain an immediate gain from an EXP file.
Houdini is defenseless against his own playing style.
The more powerful the engines, the closer we get to the theory "perfect chess is a draw"

Links :
Stockfish 15.1 vs. its own playing style
Dragon 3.3 vs. its own playing style
Houdini 6.03 Pro vs. its own playing style
Deep Rybka 4.1 vs. its own playing style

[deeds]

Creating a BIN book from a list of BIN books
For black, createBinBook creates from scratch a BIN book based on the most frequently occurring positions/moves from tens of BIN books :

For white, I prefer to use bruteforce_binbook to create a specifically optimized book

[deeds]

Openings to learn
For my 15th learning session, i use a learning farm with 4 trainings at the time :

opening #1 = (C65) 1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. d3 Bc5 5. Bxc6 dxc6 6. O-O Nd7 7. Nbd2 O-O 8. Nc4 Re8 9. Bd2 f6 10. Nh4 Bf8 *
opening #2 = (C65) 1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. d3 d6 *
opening #3 = (E05) 1. d4 Nf6 2. c4 e6 3. g3 d5 4. Bg2 dxc4 5. Nf3 Be7 6. O-O O-O 7. Qc2 a6 8. Qxc4 b5 9. Qc2 Bb7 10. Bf4 Nc6 11. Rd1 *
opening #4 = (E41) 1. Nf3 Nf6 2. c4 e6 3. Nc3 Bb4 4. e3 c5 5. d4 Bxc3+ *

The learning of the opening #1 is over :

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L  D(%)  OppAvg  OppN
   1 brainlearn 19.2 nn-e1fb1ade4432    :       0     21   250.0     500  50.0    1   498    1  99.6       0     1
   2 Aurora 301022 nn-e1fb1ade4432      :       0     22   250.0     500  50.0    1   498    1  99.6       0     1
   3 Eman experience                    :       0   ----  1002.5    2000  50.1    7  1991    2  99.5      -1     4
   4 HypnoS 051122 nn-e1fb1ade4432      :      -1     22   249.5     500  49.9    0   499    1  99.8       0     1
   5 ShashChess 25.3 nn-e1fb1ade4432    :      -3     21   248.0     500  49.6    0   496    4  99.2       0     1

White advantage = 0.18 +/- 5.62
Draw rate (equal opponents) = 50.00 % +/- 0.00

The learning of the opening #2 is over :

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L  D(%)  OppAvg  OppN
   1 HypnoS 051122 nn-e1fb1ade4432      :       1     22   250.5     500  50.1    1   499    0  99.8       0     1
   2 Eman experience                    :       0   ----  1004.0    2000  50.2   13  1982    5  99.1      -1     4
   3 ShashChess 25.3 nn-e1fb1ade4432    :      -1     22   249.5     500  49.9    1   497    2  99.4       0     1
   4 Aurora 301022 nn-e1fb1ade4432      :      -1     22   249.0     500  49.8    1   496    3  99.2       0     1
   5 brainlearn 19.2 nn-e1fb1ade4432    :      -4     21   247.0     500  49.4    2   490    8  98.0       0     1

White advantage = 0.00 +/- 5.60
Draw rate (equal opponents) = 50.00 % +/- 0.00

The learning of the opening #3 is over :

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L   D(%)  OppAvg  OppN
   1 Eman experience                    :       0   ----  1001.0    2000  50.0    2  1998    0   99.9      -0     4
   2 HypnoS 051122 nn-e1fb1ade4432      :       0     22   250.0     500  50.0    0   500    0  100.0       0     1
   3 ShashChess 25.3 nn-e1fb1ade4432    :       0     22   250.0     500  50.0    0   500    0  100.0       0     1
   4 Aurora 301022 nn-e1fb1ade4432      :       0     22   250.0     500  50.0    0   500    0  100.0       0     1
   5 brainlearn 19.2 nn-e1fb1ade4432    :      -1     21   249.0     500  49.8    0   498    2   99.6       0     1

White advantage = 0.35 +/- 5.62
Draw rate (equal opponents) = 50.00 % +/- 0.00

The learning of the opening #4 is over :

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L  D(%)  OppAvg  OppN
   1 Eman experience                    :       0   ----  1012.5    2000  50.6   36  1953   11  97.7      -4     4
   2 brainlearn 19.2 nn-e1fb1ade4432    :      -3     21   248.0     500  49.6    2   492    6  98.4       0     1
   3 HypnoS 051122 nn-e1fb1ade4432      :      -4     22   247.5     500  49.5    3   489    8  97.8       0     1
   4 Aurora 301022 nn-e1fb1ade4432      :      -5     22   246.5     500  49.3    2   489    9  97.8       0     1
   5 ShashChess 25.3 nn-e1fb1ade4432    :      -6     21   245.5     500  49.1    4   483   13  96.6       0     1

White advantage = -1.58 +/- 5.61
Draw rate (equal opponents) = 50.00 % +/- 0.00

Among these 4 openings, guess which one comes from the official [ECO CODES] opening list ?

Answer : the opening #1 comes from the [MZ V1 10Moves] opening list, the opening #2 comes from the [HERT 500] opening list, the opening #3 comes from the [ECO CODES] opening list and the opening #4 comes from the [BALSA V500] opening list.

[deeds]

Openings to learn
For my 15th learning session, i use a learning farm with 4 trainings at the time :

opening #1 = (B78) 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 g6 6. Be3 Bg7 7. f3 O-O 8. Qd2 Nc6 9. Bc4 Bd7 10. O-O-O Rc8 11. Bb3 Ne5 12. Kb1 Nc4 13. Bxc4 Rxc4 14. g4 b5 *
opening #2 = (B99) 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Be7 8. Qf3 Qc7 9. O-O-O Nbd7 10. g4 b5 11. Bxf6 Nxf6 12. g5 Nd7 13. f5 Bxg5+ 14. Kb1 Ne5 *
opening #3 = (B99) 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Bg5 e6 7. f4 Be7 8. Qf3 Qc7 9. O-O-O Nbd7 10. Be2 b5 11. Bxf6 Nxf6 12. e5 Bb7 13. Qg3 dxe5 14. fxe5 Nd7 15. Bxb5 *
opening #4 = (D30) 1. d4 d5 2. c4 c6 3. Nf3 e6 4. e3 Nf6 5. Qc2 Bd6 *