Page 22 of 40

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

Posted: Mon Aug 21, 2023 5:01 am
by deeds
Overall :

Code: Select all

   # PLAYER                            :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_190623.exp (Eman 8.40)     :      53     27   183.5     320  57.3   56  255    9  79.7       0     1
   2 Chucaro 4 UPD2.bin (Eman 8.40)    :       0   ----   136.5     320  42.7    9  255   56  79.7      53     1

White advantage = 39.35 +/- 13.64
Draw rate (equal opponents) = 50.00 % +/- 0.00

Code: Select all

002 players                : avg. D32, 002 221 ms/move, 000 320 games, 00 029 923 moves

chucaro 4 upd2 (eman 8.40) : avg. D33, 002 458 ms/move, 000 320 games, 00 013 507 moves
depth4_190623 (eman 8.40)  : avg. D32, 002 026 ms/move, 000 320 games, 00 016 416 moves

Averages :
0 208 sec/game (3 min/game)
111 plies/game (094 played plies/game)

Game stages :
- opening @ avg. D27, 003 619 ms/move
- middleg @ avg. D30, 001 813 ms/move
- endgame @ avg. D40, 001 267 ms/move

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

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

Image

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.

Re: training mode (Eman) with trouney mode (Raptora)

Posted: Mon Aug 21, 2023 2:24 pm
by deeds
deeds wrote: Fri Jul 28, 2023 4:43 pm

Code: Select all

   # PLAYER                              :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Eman_260723.exp (Eman 8.40)         :      19     28   158.0     300  52.7   29  258   13  86.0       0     1
   2 Raptora_250723.exp (RapTora 2.2)    :       0   ----   142.0     300  47.3   13  258   29  86.0      19     1

White advantage = 44.76 +/- 13.88
Draw rate (equal opponents) = 50.00 % +/- 0.00
Sorry guys, it's true that i forgot to show the usual stats :oops:

Code: Select all

002 players                  : avg. D32, 001 746 ms/move, 000 300 games, 00 031 884 moves

eman_260723 (eman 8.40)      : avg. D31, 001 992 ms/move, 000 300 games, 00 015 946 moves
raptora_250723 (raptora 2.2) : avg. D34, 001 500 ms/move, 000 300 games, 00 015 938 moves

Averages :
0 186 sec/game (3 min/game)
128 plies/game (106 played plies/game)

Game stages :
- opening @ avg. D26, 002 452 ms/move
- middleg @ avg. D30, 001 596 ms/move
- endgame @ avg. D41, 001 232 ms/move
Thanks guys for noticing it ;)

Let's take the opportunity to say a few more words :

- As already explained, the Raptora's default settings use Raptora.exp as an opening book so Raptora saved some seconds, had fewer moves to play and got a greater average depth.

- All this was useless against an engine on training mode which learned at each round and started to play with a pre-trained experience file (expRate from 4% to 6%).

Experience Book Eval Importance

Posted: Fri Aug 25, 2023 6:52 am
by deeds
10% = setoption name Experience Book Eval Importance value 1
20% = setoption name Experience Book Eval Importance value 2
...


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 nn-e1fb1ade4432.nnue net
experience data : Depth4_180423.exp (Experience Read Only true, Experience Book true, Experience Book Min Depth 27, Experience Book Max Moves 100)

opening list : startpos

Code: Select all

   # PLAYER                                   :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 20%-D27 Depth4_180423.exp (Eman 8.40)    :       8     62    59.0     112  52.7    6  106    0  94.6     -11     8
   2 10%-D27 Depth4_180423.exp (Eman 8.40)    :       6     61    58.5     112  52.2    5  107    0  95.5     -10     8
   3 40%-D27 Depth4_180423.exp (Eman 8.40)    :       3     64    58.0     112  51.8    4  108    0  96.4     -10     8
   4 30%-D27 Depth4_180423.exp (Eman 8.40)    :       3     63    58.0     112  51.8    4  108    0  96.4     -10     8
   5 50%-D27 Depth4_180423.exp (Eman 8.40)    :       0   ----    57.5     112  51.3    3  109    0  97.3     -10     8 <= Khalid's default settings for tourneys
   6 60%-D27 Depth4_180423.exp (Eman 8.40)    :       0     63    57.5     112  51.3    3  109    0  97.3     -10     8
   7 80%-D27 Depth4_180423.exp (Eman 8.40)    :      -6     61    56.5     112  50.4    4  105    3  93.8      -9     8
   8 70%-D27 Depth4_180423.exp (Eman 8.40)    :      -8     64    56.0     112  50.0    5  102    5  91.1      -9     8
   9 90%-D27 Depth4_180423.exp (Eman 8.40)    :     -83     66    43.0     112  38.4    0   86   26  76.8       1     8

White advantage = 23.98 +/- 11.06
Draw rate (equal opponents) = 50.00 % +/- 0.00

Code: Select all

009 players                       : avg. D35, 001 417 ms/move, 000 504 games, 00 045 589 moves

10%-d27 depth4_180423 (eman 8.40) : avg. D35, 001 384 ms/move, 000 112 games, 00 005 113 moves
20%-d27 depth4_180423 (eman 8.40) : avg. D36, 001 380 ms/move, 000 112 games, 00 005 170 moves
30%-d27 depth4_180423 (eman 8.40) : avg. D37, 001 444 ms/move, 000 112 games, 00 004 700 moves
40%-d27 depth4_180423 (eman 8.40) : avg. D38, 001 415 ms/move, 000 112 games, 00 004 949 moves
50%-d27 depth4_180423 (eman 8.40) : avg. D38, 001 474 ms/move, 000 112 games, 00 004 697 moves <= Khalid's default settings for tourneys
60%-d27 depth4_180423 (eman 8.40) : avg. D36, 001 459 ms/move, 000 112 games, 00 004 956 moves
70%-d27 depth4_180423 (eman 8.40) : avg. D34, 001 460 ms/move, 000 112 games, 00 004 875 moves
80%-d27 depth4_180423 (eman 8.40) : avg. D35, 001 474 ms/move, 000 112 games, 00 004 696 moves
90%-d27 depth4_180423 (eman 8.40) : avg. D29, 001 305 ms/move, 000 112 games, 00 006 433 moves

Averages :
0 128 sec/game (2 min/game)
094 plies/game (090 played plies/game)

Game stages :
- opening @ avg. D37, 001 046 ms/move
- middleg @ avg. D30, 001 804 ms/move
- endgame @ avg. D38, 001 449 ms/move
link : games
only for family, friends and customers (request password and information by PM)

Re: Experience Book Eval Importance

Posted: Fri Aug 25, 2023 7:38 am
by deeds
CONFORMITY WITH EXPERIENCE

10%-d27 depth4_180423.exp (eman 8.40) : 87% (1469 moves), longest = 18 moves
20%-d27 depth4_180423.exp (eman 8.40) : 85% (1453 moves), longest = 18 moves
30%-d27 depth4_180423.exp (eman 8.40) : 86% (1594 moves), longest = 16 moves
40%-d27 depth4_180423.exp (eman 8.40) : 77% (1094 moves), longest = 18 moves
50%-d27 depth4_180423.exp (eman 8.40) : 78% (1094 moves), longest = 19 moves <= Khalid's default settings for tourneys
60%-d27 depth4_180423.exp (eman 8.40) : 80% (1221 moves), longest = 19 moves
70%-d27 depth4_180423.exp (eman 8.40) : 75% (959 moves), longest = 9 moves
80%-d27 depth4_180423.exp (eman 8.40) : 67% (713 moves), longest = 7 moves
90%-d27 depth4_180423.exp (eman 8.40) : 66% (679 moves), longest = 3 moves

Re: merged vs pure experience data

Posted: Sat Aug 26, 2023 4:23 am
by EmptikBest
So by "merged" experience data you mean data from different engines right?

Is it fine if I merge experience data from the SAME eman engine just a different instance of it? (Since I run 7 instances of 2 threads in LearningFarm)

Re: merged vs pure experience data

Posted: Sat Aug 26, 2023 5:15 am
by deeds
EmptikBest wrote: Sat Aug 26, 2023 4:23 am So by "merged" experience data you mean data from different engines right?
Yes as explained here (by me) and confirmed there (by Sarona).

EmptikBest wrote: Sat Aug 26, 2023 4:23 am Is it fine if I merge experience data from the SAME eman engine just a different instance of it? (Since I run 7 instances of 2 threads in LearningFarm)
Yes if you understand what it does.

For example, if you merge 2 experience files from the same engine's release but with different nets and/or different settings :arrow: the scores will be averaged, it will keep the greatest depthes, etc.

Usual errors :
- to merge experience data containing pawn-based scores and wdl-based scores (see MZ, Cubail, etc.)
- to merge experience data containing pawn-based or wdl-based scores and Self Qlearning-based scores (see customers of AM's book/experience.bin file)
- etc

Re: merged vs pure experience data

Posted: Sat Aug 26, 2023 5:27 am
by EmptikBest
deeds wrote: Sat Aug 26, 2023 5:15 am
EmptikBest wrote: Sat Aug 26, 2023 4:23 am So by "merged" experience data you mean data from different engines right?
Yes as explained here (by me) and confirmed there (by Sarona).

EmptikBest wrote: Sat Aug 26, 2023 4:23 am Is it fine if I merge experience data from the SAME eman engine just a different instance of it? (Since I run 7 instances of 2 threads in LearningFarm)
Yes if you understand what it does.

For example, if you merge 2 experience files from the same engine's release but with different nets and/or different settings :arrow: the scores will be averaged, it will keep the greatest depthes, etc.

Usual errors :
- to merge experience data containing pawn-based scores and wdl-based scores (see MZ, Cubail, etc.)
- to merge experience data containing pawn-based or wdl-based scores and Self Qlearning-based scores (see customers of AM's book/experience.bin file)
- etc
Okay, thanks! It seems I have not made any of those mistakes.. I am using same net and same settings on all instances, and they all are Eman v8.91

Re: Experience Book Eval Importance

Posted: Tue Aug 29, 2023 6:29 am
by botunnet
10% = setoption name Experience Book Eval Importance value 1
20% = setoption name Experience Book Eval Importance value 2
etc.

50%-D27... <= Khalid's default settings for tourneys

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 nn-e1fb1ade4432.nnue net
experience data : Rulleeeee_030523.exp (Experience Read Only true, Experience Book true, Experience Book Min Depth 27, Experience Book Max Moves 100)

opening list : startpos

ORDO

Code: Select all

   # PLAYER                                      :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L   D(%)  OppAvg  OppN
   1 10%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    181     8.0      16  50.0    0   16    0  100.0       0     8
   2 70%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    178     8.0      16  50.0    0   16    0  100.0       0     8
   3 30%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    173     8.0      16  50.0    0   16    0  100.0       0     8
   4 50%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0   ----     8.0      16  50.0    0   16    0  100.0       0     8
   5 40%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    173     8.0      16  50.0    0   16    0  100.0       0     8
   6 90%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    180     8.0      16  50.0    0   16    0  100.0       0     8
   7 80%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    172     8.0      16  50.0    0   16    0  100.0       0     8
   8 20%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    175     8.0      16  50.0    0   16    0  100.0       0     8
   9 60%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    178     8.0      16  50.0    0   16    0  100.0       0     8

White advantage = 0.00 +/- 30.99
Draw rate (equal opponents) = 50.00 % +/- 0.00

STATISTICS

Code: Select all

009 players                          : avg. D40, 001 384 ms/move, 000 072 games, 00 005 687 moves

10%-d27 rulleeeee_030523 (eman 8.40) : avg. D39, 001 396 ms/move, 000 016 games, 00 000 621 moves
20%-d27 rulleeeee_030523 (eman 8.40) : avg. D40, 001 377 ms/move, 000 016 games, 00 000 621 moves
30%-d27 rulleeeee_030523 (eman 8.40) : avg. D39, 001 453 ms/move, 000 016 games, 00 000 662 moves
40%-d27 rulleeeee_030523 (eman 8.40) : avg. D39, 001 449 ms/move, 000 016 games, 00 000 593 moves
50%-d27 rulleeeee_030523 (eman 8.40) : avg. D44, 001 122 ms/move, 000 016 games, 00 000 616 moves
60%-d27 rulleeeee_030523 (eman 8.40) : avg. D42, 001 266 ms/move, 000 016 games, 00 000 658 moves
70%-d27 rulleeeee_030523 (eman 8.40) : avg. D40, 001 435 ms/move, 000 016 games, 00 000 699 moves
80%-d27 rulleeeee_030523 (eman 8.40) : avg. D40, 001 479 ms/move, 000 016 games, 00 000 639 moves
90%-d27 rulleeeee_030523 (eman 8.40) : avg. D38, 001 478 ms/move, 000 016 games, 00 000 578 moves

Averages :
0 109 sec/game (2 min/game)
081 plies/game (079 played plies/game)

Game stages :
- opening @ avg. D44, 000 447 ms/move
- middleg @ avg. D34, 001 910 ms/move
- endgame @ avg. D41, 001 845 ms/move

CONFORMITY WITH EXPERIENCE

10%-d27 rulleeeee_030523.exp (eman 8.40) : 96% (105 moves), longest = 8 moves
20%-d27 rulleeeee_030523.exp (eman 8.40) : 96% (105 moves), longest = 8 moves
30%-d27 rulleeeee_030523.exp (eman 8.40) : 96% (105 moves), longest = 8 moves
40%-d27 rulleeeee_030523.exp (eman 8.40) : 100% (88 moves), longest = 17 moves
50%-d27 rulleeeee_030523.exp (eman 8.40) : 91% (140 moves), longest = 12 moves
60%-d27 rulleeeee_030523.exp (eman 8.40) : 84% (95 moves), longest = 12 moves
70%-d27 rulleeeee_030523.exp (eman 8.40) : 86% (122 moves), longest = 17 moves
80%-d27 rulleeeee_030523.exp (eman 8.40) : 84% (116 moves), longest = 13 moves
90%-d27 rulleeeee_030523.exp (eman 8.40) : 62% (74 moves), longest = 13 moves

link : games
only for family, friends and customers (request password and information by PM)

Re: Experience Book Eval Importance

Posted: Tue Aug 29, 2023 6:40 am
by deeds
10% = setoption name Experience Book Eval Importance value 1
20% = setoption name Experience Book Eval Importance value 2
etc.

50%-D27... <= Khalid's default settings for tourneys

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 nn-e1fb1ade4432.nnue net
experience data : Rulleeeee_030523.exp (Experience Read Only true, Experience Book true, Experience Book Min Depth 27, Experience Book Max Moves 100)

opening list : startpos

ORDO

Code: Select all

   # PLAYER                                      :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L   D(%)  OppAvg  OppN
   1 10%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    181     8.0      16  50.0    0   16    0  100.0       0     8
   2 70%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    178     8.0      16  50.0    0   16    0  100.0       0     8
   3 30%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    173     8.0      16  50.0    0   16    0  100.0       0     8
   4 50%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0   ----     8.0      16  50.0    0   16    0  100.0       0     8
   5 40%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    173     8.0      16  50.0    0   16    0  100.0       0     8
   6 90%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    180     8.0      16  50.0    0   16    0  100.0       0     8
   7 80%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    172     8.0      16  50.0    0   16    0  100.0       0     8
   8 20%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    175     8.0      16  50.0    0   16    0  100.0       0     8
   9 60%-D27 Rulleeeee_030523.exp (Eman 8.40)    :       0    178     8.0      16  50.0    0   16    0  100.0       0     8

White advantage = 0.00 +/- 30.99
Draw rate (equal opponents) = 50.00 % +/- 0.00

STATISTICS

Code: Select all

009 players                          : avg. D40, 001 384 ms/move, 000 072 games, 00 005 687 moves

10%-d27 rulleeeee_030523 (eman 8.40) : avg. D39, 001 396 ms/move, 000 016 games, 00 000 621 moves
20%-d27 rulleeeee_030523 (eman 8.40) : avg. D40, 001 377 ms/move, 000 016 games, 00 000 621 moves
30%-d27 rulleeeee_030523 (eman 8.40) : avg. D39, 001 453 ms/move, 000 016 games, 00 000 662 moves
40%-d27 rulleeeee_030523 (eman 8.40) : avg. D39, 001 449 ms/move, 000 016 games, 00 000 593 moves
50%-d27 rulleeeee_030523 (eman 8.40) : avg. D44, 001 122 ms/move, 000 016 games, 00 000 616 moves
60%-d27 rulleeeee_030523 (eman 8.40) : avg. D42, 001 266 ms/move, 000 016 games, 00 000 658 moves
70%-d27 rulleeeee_030523 (eman 8.40) : avg. D40, 001 435 ms/move, 000 016 games, 00 000 699 moves
80%-d27 rulleeeee_030523 (eman 8.40) : avg. D40, 001 479 ms/move, 000 016 games, 00 000 639 moves
90%-d27 rulleeeee_030523 (eman 8.40) : avg. D38, 001 478 ms/move, 000 016 games, 00 000 578 moves

Averages :
0 109 sec/game (2 min/game)
081 plies/game (079 played plies/game)

Game stages :
- opening @ avg. D44, 000 447 ms/move
- middleg @ avg. D34, 001 910 ms/move
- endgame @ avg. D41, 001 845 ms/move

CONFORMITY WITH EXPERIENCE

10%-d27 rulleeeee_030523.exp (eman 8.40) : 96% (105 moves), longest = 8 moves
20%-d27 rulleeeee_030523.exp (eman 8.40) : 96% (105 moves), longest = 8 moves
30%-d27 rulleeeee_030523.exp (eman 8.40) : 96% (105 moves), longest = 8 moves
40%-d27 rulleeeee_030523.exp (eman 8.40) : 100% (88 moves), longest = 17 moves
50%-d27 rulleeeee_030523.exp (eman 8.40) : 91% (140 moves), longest = 12 moves
60%-d27 rulleeeee_030523.exp (eman 8.40) : 84% (95 moves), longest = 12 moves
70%-d27 rulleeeee_030523.exp (eman 8.40) : 86% (122 moves), longest = 17 moves
80%-d27 rulleeeee_030523.exp (eman 8.40) : 84% (116 moves), longest = 13 moves
90%-d27 rulleeeee_030523.exp (eman 8.40) : 62% (74 moves), longest = 13 moves

link : games
only for family, friends and customers (request password and information by PM)