Sedat Canbaz wrote:
Where Depth4 2015 book is tuned to Prefer both Depth 2025's Exp series..
Deeds, who seems to be Responsible for all the openings (played by Depth4..)
source
To get a good variety of experience data, I train Eman on several opening lists with several opening books as sparring-partners :
How it works ?
For each opening of each opening list, the GUI (arena, cutechess, etc.) plays the opening moves then the engines start to play from the last position.
Let's take a training game played from the "1.b3 Nimzovich-Larsen" opening of the "20 first moves" list :
[Event "EXP VS BOOK"]
[Site "dual xeon e5-2660v3"]
[Date "2025.02.12"]
[Round "1"]
[White "Depth4_120824 (Eman 8.40)"]
[Black "Brain_110323 (Eman 8.40)"]
[Result "1/2-1/2"]
[ECO "A01"]
[GameDuration "00:02:48"]
[Opening "Nimzovich-Larsen attack"]
[PlyCount "55"]
[TimeControl "60+1"]
[Variation "Classical Variation"]
1. b3 {book} d5 2. Nf3 {0.00/29 4.1s} Bf5 3. Bb2 {-0.09/22 1.0s} e6
4. c4 {-0.05/29 10s} c6 {-0.13/31 12s} 5. e3 {-0.04/29 4.1s} Nd7 {-0.10/33 3.9s}
6. Nc3 {-0.08/26 1.8s} Ngf6 {-0.01/31 6.3s} 7. Nh4 {-0.07/26 1.5s}
Bg4 {-0.05/33 2.9s} 8. Be2 {-0.13/25 1.1s} Bxe2 {-0.02/30 1.0s}
9. Qxe2 {-0.15/27 5.1s} Nc5 {-0.02/31 2.1s} 10. cxd5 {-0.04/26 0.81s}
Nxd5 {-0.02/30 2.2s} 11. Nf3 {0.00/27 1.4s} Nxc3 {0.00/30 1.9s}
12. Bxc3 {+0.05/27 1.1s} f6 {0.00/28 1.7s} 13. Qc4 {+0.16/23 1.5s}
Qd5 {0.00/26 1.6s} 14. Ke2 {+0.01/23 1.3s} O-O-O {+0.13/20 1.3s}
15. g4 {0.00/24 2.6s} Qd7 {0.00/29 1.3s} 16. Bd4 {0.00/36 1.9s}
Kb8 {0.00/38 1.4s} 17. g5 {0.00/39 3.6s} e5 {0.00/46 1.9s}
18. Bxc5 {0.00/44 1.2s} Bxc5 {0.00/46 1.6s} 19. Qxc5 {0.00/48 3.4s}
Qd3+ {0.00/48 1.8s} 20. Kd1 {0.00/50 1.7s} Qe4 {0.00/48 1.6s}
21. Ke2 {0.00/53 3.7s} Qd3+ {0.00/52 10.0s} 22. Ke1 {0.00/55 5.2s}
Qe4 {0.00/56 2.4s} 23. gxf6 {0.00/58 16s} gxf6 {0.00/56 2.4s}
24. Ke2 {0.00/61 1.7s} Qd3+ {0.00/56 1.6s} 25. Kd1 {0.00/57 3.4s}
Qe4 {0.00/58 2.1s} 26. Ke2 {0.00/60 0.49s} Qd3+ {0.00/60 13s}
27. Kd1 {0.00/62 2.8s} Qe4 {0.00/64 4.9s}
28. Ke2 {0.00/61 0.64s, Draw by 3-fold repetition} 1/2-1/2
In this training game,
- the move "1. b3 {book}" is played by the GUI,
- the move "1. ... d5" comes from the "Brain_110323" BIN book,
- the move "2. Nf3 {0.00/29 4.1s}" is evaluated by the "Eman 8.40" engine which will update the "Depth4_120824" EXP file,
- the move "4. ... c6 {-0.13/31 12s}" is evaluated by the "Eman 8.40" engine because the "Brain_110323" BIN book is out-of-book,
-etc.
- no move played as white comes from the "Depth4_120824" EXP file because at training the "experience book" option is set to false. The moves are evaluated by the "Eman 8.40" engine based on the missing experience data in the "Depth4_120824" EXP file,
- only the moves played as black without evaluation represent the "Brain_110323" BIN book,
In sum :
- training games don't represent the accuracy of the EXP file,
- opening moves played by the GUI don't represent the EXP file,
- moves evaluated by the engine don't represent the EXP file,
- If a clown creates an opening book from training games, he alone is responsible for the good or bad performance of his own opening book.
his experience files have not learned any of the openings he used for his small matches 
