Train your engine !

[ZamChess]

After a few tries, I discovered that the following settings allowed better learning :

Thank you from the bottom of my heart for sharing with us the results and observations of your tests.
All the best.

[deeds]

Thank you from the bottom of my heart for sharing with us the results and observations of your tests.
All the best.

Hi ZamChess, I found other settings for training that are even more effective :

Experience Book Min Depth = 4 and Experience Book Eval Importance = 1
In very rare cases (1), Experience Book Min Depth = 40 + Experience Book Eval Importance = 0 may cause Eman to lose some games unnecessarily.
With the new settings, training remains effective even after several hundred games.

Automatic Time Management = true
After a few dozen games, Eman uses the moves from the experience file.
At the start of the game, Eman saves time and even wins time.
We can therefore reactivate this function so that Eman can think for a long time in strategic positions or when there are several equivalent moves in his experience file.

ENJOY !

(1) For the same position, if the experiment file contains:
- a move # 1 which was very effective and whose depth> = D40
- a move # 2 which was less efficient and whose depth <D40
In this case, Eman will still play move # 1 even if it results in further defeats (= his efficiency drops).

[deeds]

B06 Modern :

[Event "Eman's learning"]
[Site "deeds"]
[Date "2021.05.24"]
[Round "0195-0196"]
[White "B06"]
[Black "B06"]
[Result "*"]
[ECO "B06"]
[PlyCount "20"]

1. e4 d6 2. d4 g6 3. Nc3 Bg7 4. Bg5 a6 5. Bc4 Nc6 6. Nf3 h6 7. Be3 e6 8.
Qe2 Nf6 9. O-O-O O-O 10. e5 Nd7 *

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L  D(%)  OppAvg  OppN
   1 Eman experience                    :       3     11  1008.0    2000  50.4   56  1904   40  95.2       0     1
   2 brainlearn 13.1 nn-cdf1785602d6    :       0   ----   992.0    2000  49.6   40  1904   56  95.2       3     1

White advantage = 16.84 +/- 5.61
Draw rate (equal opponents) = 50.00 % +/- 0.00

brainlearn 13.1 nn-cdf1785602d6 with 40/2000 wins (2,00%, D27/245, 1 482 ms) :

1. e4 @ 100,00%

Eman experience with 56/2000 wins (2,80%, D44/245, 1 242 ms) :

1. e4 @ 100,00%

Averages :
135 sec/game
120 plies/game

Games from 1 to 1200 were played with :
- Experience Book Min Depth = 40
- Experience Book Eval Importance = 0
- Automatic Time Management = false

Games from 1201 to 1800 were played with :
- Experience Book Min Depth = 4
- Experience Book Eval Importance = 1
- Automatic Time Management = false

Games from 1801 to 2000 were played with :
- Experience Book Min Depth = 4
- Experience Book Eval Importance = 1
- Automatic Time Management = true

The red curve goes up at the end.

[deeds]

D17 Slave, Central :

[Event "Eman's learning"]
[Site "deeds"]
[Date "2021.05.24"]
[Round "003233"]
[White "eman"]
[Black "eman"]
[Result "*"]
[ECO "D17"]
[PlyCount "28"]

1. d4 d5 2. c4 c6 3. Nf3 Nf6 4. Nc3 dxc4 5. a4 Bf5 6. Ne5 e6 7. f3 Bb4 8.
e4 Bxe4 9. fxe4 Nxe4 10. Bd2 Qxd4 11. Nxe4 Qxe4+ 12. Qe2 Bxd2+ 13. Kxd2
Qd5+ 14. Kc2 Na6 *

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L  D(%)  OppAvg  OppN
   1 brainlearn 13.1 nn-cdf1785602d6    :       0   ----  1014.0    2000  50.6   32  1962    6  98.1      -5     1
   2 Eman experience                    :      -5     11   988.0    2000  49.4    6  1962   32  98.1       0     1

White advantage = 6.65 +/- 5.62
Draw rate (equal opponents) = 50.00 % +/- 0.00

brainlearn 13.1 nn-cdf1785602d6 with 32/2000 wins (1,60%, D27/245, 1 487 ms) :

1. d4 @ 100,00%

Eman experience with 6/2000 wins (0,30%, D40/245, 1 239 ms) :

1. d4 @ 100,00%

Averages :
107 sec/game
108 plies/game

Games from 1 to 1150 were played with :
- Experience Book Min Depth = 4
- Experience Book Eval Importance = 1
- Automatic Time Management = true

Games from 1151 to 2000 were played with :
- (b]Experience Book Min Depth = 50[/b]
- Experience Book Eval Importance = 1
- Automatic Time Management = false

The red curve stabilized then started to rise very slightly. As there are very few moves with D50 and +higher in the experience file, Eman will not save much time so i have to deactivate Automatic Time Management (which consumes too much time in the first moves then stresses the engine at the end of the game and brings too many unnecessary losses).

[deeds]

D89 Grunfeld, Classical Exchange, Sokolsky :

[Event "Eman's learning"]
[Site "deeds"]
[Date "2021.05.24"]
[Round "003831"]
[White "eman"]
[Black "eman"]
[Result "*"]
[ECO "D89"]
[PlyCount "28"]

1. d4 Nf6 2. c4 g6 3. Nc3 d5 4. cxd5 Nxd5 5. e4 Nxc3 6. bxc3 Bg7 7. Bc4
O-O 8. Ne2 c5 9. O-O Nc6 10. Be3 cxd4 11. cxd4 Bg4 12. f3 Na5 13. Bd3 Be6
14. Rc1 Bxa2 *

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L   D(%)  OppAvg  OppN
   1 Eman experience                    :       0     11  1000.5    2000  50.0    1  1999    0  100.0       0     1
   2 brainlearn 13.1 nn-ac07bd334b62    :       0   ----   999.5    2000  50.0    0  1999    1  100.0       0     1

White advantage = 0.18 +/- 5.61
Draw rate (equal opponents) = 50.00 % +/- 0.00

brainlearn 13.1 nn-ac07bd334b62 with 0/2000 wins (0.00%, D33/245, 1 319 ms) :

1. d4 @ 0.00%

Eman experience with 1/2000 wins (0,05%, D77/245, 1 021 ms) :

1. d4 @ 100,00%

Averages :
78 sec/game
95 plies/game

Here i updated the net for both engines.
It's funny to see the only win in 2000 games was by Eman loool

The last 500 games were played with :
- Experience Book Min Depth = 60
Min D60 in order to vary the first Eman's moves but it didn't find a winning way here.

[deeds]

D89 Grunfeld, Classical Exchange :

[Event "Eman's learning"]
[Site "deeds"]
[Date "2021.05.24"]
[Round "003832"]
[White "eman"]
[Black "eman"]
[Result "*"]
[ECO "D89"]
[PlyCount "28"]

1. d4 Nf6 2. c4 g6 3. Nc3 d5 4. cxd5 Nxd5 5. e4 Nxc3 6. bxc3 Bg7 7. Bc4
O-O 8. Ne2 c5 9. O-O Nc6 10. Be3 cxd4 11. cxd4 Bg4 12. f3 Na5 13. Bd3 Be6
14. d5 Bxa1 *

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L  D(%)  OppAvg  OppN
   1 brainlearn 13.1 nn-ac07bd334b62    :       0   ----  1000.5    2000  50.0    2  1997    1  99.8      -0     1
   2 Eman experience                    :       0     11   999.5    2000  50.0    1  1997    2  99.8       0     1

White advantage = 0.18 +/- 5.63
Draw rate (equal opponents) = 50.00 % +/- 0.00

brainlearn 13.1 nn-ac07bd334b62 with 2/2000 wins (0,10%, D34/245, 1 313 ms) :

1. d4 @ 100,00%

Eman experience with 1/2000 wins (0,05%, D79/245, 1 424 ms) :

1. d4 @ 100,00%

Averages :
57 sec/game
71 plies/game

The last 500 games were played with :
- Experience Book = false
- Automatic Time Management = true
But nothing seems to help both engines here...

[ZamChess]

Thank you very much for your tests and observations.
I wish you and your family a Happy New Year.

Friendly Zam.

[deeds]

Thank you, Happy New Year !

[deeds]

A00 Grob Opening (Fric-Kilibr) :

[Event "Eman's learning"]
[Site "deeds"]
[Date "2021.05.24"]
[Round "1"]
[White "eman"]
[Black "eman"]
[Result "*"]
[ECO "A00"]
[PlyCount "1"]

1. g4 *

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Eman experience                    :     134     22  1119.5    2000  56.0  935  369  696  18.4       0     1
   2 brainlearn 13.1 nn-ac07bd334b62    :       0   ----   880.5    2000  44.0  696  369  935  18.4     134     1

White advantage = -426.65 +/- 11.00
Draw rate (equal opponents) = 50.00 % +/- 0.00

brainlearn 13.1 nn-ac07bd334b62 with 696/2000 wins (34,80%, D25/245, 1 487 ms) :

1. g4 @ 100,00%

Eman experience with 935/2000 wins (46,75%, D35/245, 1 687 ms) :

1. g4 @ 100,00%

Averages :
232 sec/game
148 plies/game

Clearly the "nn-ac07bd334b62" isn't perfect nor the stronger net ever loool

[deeds]

C18 French, Winawer, Poisoned Pawn :

[Event "Eman's learning"]
[Site "deeds"]
[Date "2021.05.24"]
[Round "002195"]
[White "eman"]
[Black "eman"]
[Result "*"]
[ECO "C18"]
[PlyCount "25"]

1. e4 e6 2. d4 d5 3. Nc3 Bb4 4. e5 c5 5. a3 Bxc3+ 6. bxc3 Ne7 7. Qg4 Qc7
8. Qxg7 Rg8 9. Qxh7 cxd4 10. Ne2 Nbc6 11. f4 Bd7 12. Qd3 dxc3 13. Nxc3 *

Code: Select all

   # PLAYER                             :  RATING  ERROR  POINTS  PLAYED   (%)    W     D    L  D(%)  OppAvg  OppN
   1 brainlearn 13.1 nn-ac07bd334b62    :       0   ----  1005.5    2000  50.3   13  1985    2  99.3      -2     1
   2 Eman experience                    :      -2     11   994.5    2000  49.7    2  1985   13  99.3       0     1

White advantage = 2.28 +/- 5.63
Draw rate (equal opponents) = 50.00 % +/- 0.00

brainlearn 13.1 nn-ac07bd334b62 with 13/2000 wins (0,65%, D31/245, 1 453 ms) :

1. e4 @ 100,00%

Eman experience with 2/2000 wins (0,10%, D44/245, 2 058 ms) :

1. e4 @ 100,00%

Averages :
109 sec/game
88 plies/game