deeds_chess_system

deeds · Post by **deeds** » Sun Feb 16, 2025 8:01 pm

pawn-value based experience file : experience data purely trained by Eman engine until 8.40 release (before Nov 5, 2022 patch)

hardware : dual xeon e5-2660v3
software : cutechess-gui 1.2.0
settings : TC 60s+1s, 40 threads, hash 4096 MB, partial syzygy 7men, book width = 2

Opening list : startpos

Code: Select all

   # PLAYER                           :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_120824.exp (Eman 8.40)    :       4     50    50.5     100  50.5    1   99    0  99.0       0     1
   2 Brain 110323.bin (Eman 8.40)     :       0   ----    49.5     100  49.5    0   99    1  99.0       4     1

White advantage = 3.51 +/- 25.14
Draw rate (equal opponents) = 50.00 % +/- 0.00

Opening list : GHPPN's 15 most popular openings (Brain 110323's Opening Rate : 60%)

Code: Select all

   # PLAYER                           :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_120824.exp (Eman 8.40)    :      12     94    15.5      30  51.7    1   29    0  96.7       0     1
   2 Brain 110323.bin (Eman 8.40)     :       0   ----    14.5      30  48.3    0   29    1  96.7      12     1

White advantage = 11.70 +/- 46.76
Draw rate (equal opponents) = 50.00 % +/- 0.00

Opening list : 20 first legal moves (Brain 110323's Opening Rate : 25%)

Code: Select all

   # PLAYER                           :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_120824.exp (Eman 8.40)    :      26     82    21.5      40  53.8    3   37    0  92.5       0     1
   2 Brain 110323.bin (Eman 8.40)     :       0   ----    18.5      40  46.3    0   37    3  92.5      26     1

White advantage = -26.49 +/- 41.17
Draw rate (equal opponents) = 50.00 % +/- 0.00

Opening list : Albert Silver's 25 gambit suit (Brain 110323's Opening Rate : 4%)

Code: Select all

   # PLAYER                           :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_120824.exp (Eman 8.40)    :      57     70    29.0      50  58.0   11   36    3  72.0       0     1
   2 Brain 110323.bin (Eman 8.40)     :       0   ----    21.0      50  42.0    3   36   11  72.0      57     1

White advantage = 14.40 +/- 35.25
Draw rate (equal opponents) = 50.00 % +/- 0.00

Opening list : Jeroen Noomen's 50 tcec season 22 superfinal openings (Brain 110323's Opening Rate : 0%)

Code: Select all

   # PLAYER                           :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_120824.exp (Eman 8.40)    :     170     65    69.0     100  69.0   39   60    1  60.0       0     1
   2 Brain 110323.bin (Eman 8.40)     :       0   ----    31.0     100  31.0    1   60   39  60.0     170     1

White advantage = 163.38 +/- 33.70
Draw rate (equal opponents) = 50.00 % +/- 0.00

link : games

deeds · Post by **deeds** » Sun Feb 16, 2025 8:04 pm

Overall :

Code: Select all

   # PLAYER                           :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_120824.exp (Eman 8.40)    :      57     27   185.5     320  58.0   55  261    4  81.6       0     1
   2 Brain 110323.bin (Eman 8.40)     :       0   ----   134.5     320  42.0    4  261   55  81.6      57     1

White advantage = 41.80 +/- 13.65
Draw rate (equal opponents) = 50.00 % +/- 0.00

Code: Select all

002 players               : avg. D31, 002 074 ms/move, 000 320 games, 00 032 757 moves

brain 110323 (eman 8.40)  : avg. D32, 002 159 ms/move, 000 320 games, 00 015 759 moves
depth4_120824 (eman 8.40) : avg. D31, 001 996 ms/move, 000 320 games, 00 016 998 moves

Averages :
0 212 sec/game (4 min/game)
115 plies/game (102 played plies/game)

Game stages :
- opening @ avg. D27, 003 303 ms/move
- middleg @ avg. D29, 001 712 ms/move
- endgame @ avg. D38, 001 242 ms/move

deeds · Post by **deeds** » Sun Feb 16, 2025 8:23 pm

deeds wrote: Mon Aug 21, 2023 6:47 am Eman's training with the main openings (book width = 1) of about ten books is finished :

Opening books not being my domain, I needed some stats, some orders of magnitude, to see what their main openings looked like (book width=1), etc.

Based on the above stats, I think I can continue Eman's training even with their secondary openings (book width = 2), with other books, with a newer experience file, etc.

Eman's training with the secondary openings (book width = 2) of about ten books is finished :

Based on the above stats, with "book width = 2", the books lost about ten additional games than with "book width = 1".

I'm thinking about how to continue Eman's training with other books, with a more recent experience file, etc.

deeds · Post by **deeds** » Mon Feb 17, 2025 3:07 pm

Code: Select all

   # PLAYER                                           :  RATING  ERROR  POINTS  PLAYED   (%)     W     D    L  D(%)  OppAvg  OppN
   1 Depth4_xxxxxx.exp (Eman 8.xx)                    :       0   ----  3646.0    6400  57.0  1031  5230  139  81.7     -50    20
   2 width1 - Optimus 3.1.bin (Eman 8.xx)             :     -33     27   145.0     320  45.3    15   260   45  81.3       0     1
   3 width1 - Mantis-DH1.bin (Eman 8.xx)              :     -37     26   143.5     320  44.8     7   273   40  85.3       0     1
   4 width1 - hcanfree-upd.bin (Eman 8.xx)            :     -39     27   142.5     320  44.5     7   271   42  84.7       0     1
   5 width1 - Hubble 3.1.bin (Eman 8.xx)              :     -39     28   142.5     320  44.5     8   269   43  84.1       0     1
   6 width1 - Solista-010722.bin (Eman 8.xx)          :     -39     28   142.5     320  44.5     7   271   42  84.7       0     1
   7 width2 - Cute 130523.bin (Eman 8.xx)             :     -45     28   140.0     320  43.8    13   254   53  79.4       0     1
   8 width1 - Agarta-040623.bin (Eman 8.xx)           :     -47     28   139.0     320  43.4     6   266   48  83.1       0     1
   9 width1 - Leschats14062023.bin (Eman 8.xx)        :     -49     27   138.0     320  43.1     5   266   49  83.1       0     1
  10 width1 - Olisfish-040623.bin (Eman 8.xx)         :     -50     28   137.5     320  43.0     8   259   53  80.9       0     1
  11 width2 - Sentinel 160623.bin (Eman 8.xx)         :     -51     28   137.0     320  42.8     7   260   53  81.3       0     1
  12 width1 - Canseven-140223.bin (Eman 8.xx)         :     -52     27   136.5     320  42.7     7   259   54  80.9       0     1
  13 width1 - Chucaro 4 UPD2.bin (Eman 8.xx)          :     -52     28   136.5     320  42.7     9   255   56  79.7       0     1
  14 width2 - Cani 230323.bin (Eman 8.xx)             :     -54     27   136.0     320  42.5     6   260   54  81.3       0     1
  15 width2 - Solista Attack v3.bin (Eman 8.xx)       :     -54     27   136.0     320  42.5     5   262   53  81.9       0     1
  16 width2 - NebulaRing178.bin (Eman 8.xx)           :     -57     27   134.5     320  42.0     6   257   57  80.3       0     1
  17 width2 - StudentChess 140123.bin (Eman 8.xx)     :     -57     28   134.5     320  42.0     5   259   56  80.9       0     1
  18 width2 - Brain 110323.bin (Eman 8.xx)            :     -57     28   134.5     320  42.0     4   261   55  81.6       0     1
  19 width2 - ThunderWizard 310822.bin (Eman 8.xx)    :     -58     27   134.0     320  41.9     3   262   55  81.9       0     1
  20 width2 - CallMeX 011122.bin (Eman 8.xx)          :     -62     27   132.5     320  41.4     4   257   59  80.3       0     1
  21 width2 - M13.8.bin (Eman 8.xx)                   :     -64     28   131.5     320  41.1     7   249   64  77.8       0     1

White advantage = 37.26 +/- 3.27
Draw rate (equal opponents) = 50.00 % +/- 0.00

deeds · Post by **deeds** » Mon Feb 17, 2025 10:38 pm

deeds wrote: Sun Feb 16, 2025 8:23 pm I'm thinking about how to continue Eman's training with other books, with a more recent experience file, etc.

I should avoid training Eman with opening books from the starting position because the recent opening books offer roughly the same attacks/defenses so after a few games, they become harmless for a learning engine.

maybe i should use older opening books...

Even with "width=3" I doubt that another ten opening books can improve the accuracy of my Depth4 experience file on these 4 opening lists (ghppn, 20 first moves, silver gambit suite, tcec s22 superfinal).

maybe it's time to train Eman on other opening lists (salc, noomen, yhwh, etc.)...

deeds · Post by **deeds** » Thu Feb 27, 2025 1:14 pm

hardware : dual xeon e5-2660v3
software : arena 3.5.1
settings : TC 60s+1s, 40 threads, hash 4096 MB, partial syzygy 7men, book width = 1

Opening list : antidraw_salc.pgn (53 openings)

Code: Select all

   # PLAYER                           :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_180225.exp (Eman 8.40)    :     145     63    69.5     106  65.6   44   51   11  48.1       0     1
   2 DCbook Large.bin (Eman 8.40)     :       0   ----    36.5     106  34.4   11   51   44  48.1     145     1

White advantage = 185.16 +/- 32.71
Draw rate (equal opponents) = 50.00 % +/- 0.00

link : games
pass : 59Ef9tOGT1RCBGYgpMTuzw

deeds · Post by **deeds** » Thu Feb 27, 2025 1:39 pm

older opening books (from 2005 to 2014)

other opening lists (salc, noomen, yhwh)

new experience file (Depth4_180225.exp, 750 MB)

other GUI (arena 3.5.1)

Eman's training with opening books

deeds · Post by **deeds** » Thu Mar 06, 2025 4:04 pm

update :

deeds wrote: Wed Nov 27, 2024 8:43 pm Objective :
- checking the compatibility/sanity of each EXP file
- testing which playing styles fit the best with an engine
- revealing what experience data need to be reinforced
- revealing what missing openings need to be added
- etc.

hardware : dual xeon e5-2660v3, 40 threads, hash 4 GB
software : Arena 3.5.1, TC 1m+1s, partial syzygy 7men
engine : Eman 8.40 played with the nn-e1fb1ade4432.nnue net
experience data : free experience files

tourney mode :
- Experience Read Only=true
- Experience Book=true
- Experience Book Best Move=true
- Experience Book Min Depth=27
- Experience Book Eval Importance=5
- Experience Book Max Moves=100
- Move Overhead (ms)=1000

Opening list : startpos

After 2 rounds :

Code: Select all

   # PLAYER                                         :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L   D(%)  OppAvg  OppN
   1 Torch_v3.exp (Eman 8.40)                       :      23    142     6.5      12  54.2    1   11    0   91.7      -7     6
   2 Solidor_190924.exp (Eman 8.40)                 :      23    140     6.5      12  54.2    1   11    0   91.7      -7     6
   3 ProteusSF_280722.exp (Eman 8.40)               :      23    141     6.5      12  54.2    1   11    0   91.7      -7     6
   4 Depth43_291221.exp (Eman 8.40)                 :      23    150     6.5      12  54.2    1   11    0   91.7      -7     6
   5 Deep_Rybka_4_1.exp (Eman 8.40)                 :      10     84    16.5      32  51.6    1   31    0   96.9      -1    16
   6 Houdini_6_03_Pro.exp (Eman 8.40)               :      10     87    16.5      32  51.6    1   31    0   96.9      -1    16
   7 Johny-Petrol_Hypnos_030325.exp (Eman 8.40)     :       0     83    16.0      32  50.0    0   32    0  100.0      -0    16
   8 private DeeDs_HypnoS_231124.exp (Eman 8.40)    :       0     82    16.0      32  50.0    0   32    0  100.0      -0    16
   9 Cubail_300723.exp (Eman 8.40)                  :       0     83    16.0      32  50.0    0   32    0  100.0      -0    16
  10 Cubail_250623.exp (Eman 8.40)                  :      -7    140     6.0      12  50.0    0   12    0  100.0      -7     6
  11 Komodo_Dragon_3_3.exp (Eman 8.40)              :      -7    140     6.0      12  50.0    0   12    0  100.0      -7     6
  12 Stockfish_15_1.exp (Eman 8.40)                 :      -7    140     6.0      12  50.0    0   12    0  100.0      -7     6
  13 Cubail_271124.exp (Eman 8.40)                  :      -7    140     6.0      12  50.0    0   12    0  100.0      -7     6
  14 Khalid_210122.exp (Eman 8.40)                  :      -7    142     6.0      12  50.0    0   12    0  100.0      -7     6
  15 StockfishMZ_020722.exp (Eman 8.40)             :      -7    141     6.0      12  50.0    0   12    0  100.0      -7     6
  16 Rulleeeee_030523.exp (Eman 8.40)               :      -7    143     6.0      12  50.0    0   12    0  100.0      -7     6
  17 Johny-Petrol_Iris_190324.exp (Eman 8.40)       :     -63     86    13.0      32  40.6    0   26    6   81.3       4    16

White advantage = 13.10 +/- 21.01
Draw rate (equal opponents) = 50.00 % +/- 0.00

deeds · Post by **deeds** » Thu Mar 06, 2025 4:06 pm

update :

deeds wrote: Fri Dec 13, 2024 4:19 pm Same test with D.C.S.

deeds_chess_system.ini :
Code: Select all
playing_engine = E:\JEUX\ARENA CHESS 3.5.1\Engines\Eman\20T Eman 8.40 x64 BMI2.exe
book_engine = E:\JEUX\ARENA CHESS 3.5.1\Engines\asmFish\20T asmFish 291118 x64 BMI2.exe
log = False
tourney mode :
- Move Overhead (ms)=1000
- OwnBook=false
- BookFile=<empty>
- Experience Read Only=true
- Experience Min Depth=4
- Experience Min Count=1
- DCS Concurrent EXP=false

Opening list : startpos

After 2 rounds :

Code: Select all

   # PLAYER                                      :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L   D(%)  OppAvg  OppN
   1 Solidor_190924.exp (d.c.s.)                 :      10     82    16.5      32  51.6    1   31    0   96.9      -1    16
   2 ProteusSF_280722.exp (d.c.s.)               :      10     83    17.0      33  51.5    1   32    0   97.0      -1    16
   3 Komodo_Dragon_3_3.exp (d.c.s.)              :       0     81    16.5      33  50.0    1   31    1   93.9       0    16
   4 Cubail_271124.exp (d.c.s.)                  :       0     81    16.5      33  50.0    1   31    1   93.9      -0    16
   5 Rulleeeee_030523.exp (d.c.s.)               :       0     82    16.5      33  50.0    0   33    0  100.0      -0    16
   6 Khalid_210122.exp (d.c.s.)                  :       0     84    16.5      33  50.0    0   33    0  100.0      -0    16
   7 Torch_v3.exp (d.c.s.)                       :       0     81    16.5      33  50.0    0   33    0  100.0      -0    16
   8 Stockfish_15_1.exp (d.c.s.)                 :       0     84    16.5      33  50.0    0   33    0  100.0      -0    16
   9 Houdini_6_03_Pro.exp (d.c.s.)               :       0     84    16.0      32  50.0    0   32    0  100.0      -0    16
  10 Johny-Petrol_Iris_190324.exp (d.c.s.)       :       0     84    16.0      32  50.0    0   32    0  100.0      -0    16
  11 private DeeDs_HypnoS_231124.exp (d.c.s.)    :       0     86    16.0      32  50.0    0   32    0  100.0      -0    16
  12 Deep_Rybka_4_1.exp (d.c.s.)                 :       0     83    16.0      32  50.0    0   32    0  100.0      -0    16
  13 Cubail_300723.exp (d.c.s.)                  :       0     83    16.0      32  50.0    0   32    0  100.0      -0    16
  14 StockfishMZ_020722.exp (d.c.s.)             :       0     85    16.5      33  50.0    0   33    0  100.0       0    16
  15 Cubail_250623.exp (d.c.s.)                  :       0     77    16.5      33  50.0    0   33    0  100.0       0    16
  16 Depth43_291221.exp (d.c.s.)                 :       0     82    16.5      33  50.0    0   33    0  100.0       0    16
  17 Johny-Petrol_Hypnos_030325.exp (d.c.s.)     :     -21     86    15.0      32  46.9    0   30    2   93.8       1    16

White advantage = 2.50 +/- 15.28
Draw rate (equal opponents) = 50.00 % +/- 0.00

Anton101 · Post by **Anton101** » Sun Mar 09, 2025 10:14 pm

And where did this archive of experience come from?

Who is Johny-Petrol?

OpenChess

OpenChess

deeds_chess_system

Re: pawn-value based exp file vs polyglot bin book

Re: pawn-value based exp file vs polyglot bin book

Re: pawn-value based exp file vs polyglot bin book

Re: pawn-value based exp file vs polyglot bin book

Re: pawn-value based exp file vs polyglot bin book

Eman's training with opening books

Re: pawn-value based exp file vs polyglot bin book

Re: free exp files : sanity check

Re: free exp files : sanity check

Re: free exp files : sanity check