merged vs pure experience data

[deeds]

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

Code: Select all

Depth4_050522.exp    : experience data trained only with Eman
ProteusSF_030622.exp : experience data merged from Sarona's engine and Igor's data

Opening list : GHPPN's 15 most popular openings

Code: Select all

   # PLAYER                              :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_050522.exp (Eman 8.10)       :      12     91    15.5      30  51.7    1   29    0  96.7       0     1
   2 ProteusSF_030622.exp (Eman 8.10)    :       0   ----    14.5      30  48.3    0   29    1  96.7      12     1

White advantage = -11.70 +/- 47.15
Draw rate (equal opponents) = 50.00 % +/- 0.00

link : games
pass : ilUjtD1BLyiiL3gGfA9ihA

[deeds]

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

Code: Select all

Depth4_050522.exp    : experience data trained only with Eman
ProteusSF_030622.exp : experience data merged from Sarona's engine and Igor's data

Opening list : Albert Silver's 25 gambit suit

Code: Select all

   # PLAYER                              :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_050522.exp (Eman 8.10)       :      35     69    27.5      50  55.0    9   37    4  74.0       0     1
   2 ProteusSF_030622.exp (Eman 8.10)    :       0   ----    22.5      50  45.0    4   37    9  74.0      35     1

White advantage = 7.08 +/- 35.50
Draw rate (equal opponents) = 50.00 % +/- 0.00

link : games
pass : ilUjtD1BLyiiL3gGfA9ihA

[deeds]

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

Code: Select all

Depth4_050522.exp    : experience data trained only with Eman
ProteusSF_030622.exp : experience data merged from Sarona's engine and Igor's data

Opening list : DeeDs 50 openings list #1

Code: Select all

   # PLAYER                              :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_050522.exp (Eman 8.10)       :      32     50    54.5     100  54.5    9   91    0  91.0       0     1
   2 ProteusSF_030622.exp (Eman 8.10)    :       0   ----    45.5     100  45.5    0   91    9  91.0      32     1

White advantage = -17.68 +/- 25.31
Draw rate (equal opponents) = 50.00 % +/- 0.00

link : games
pass : ilUjtD1BLyiiL3gGfA9ihA

[deeds]

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

Code: Select all

Depth4_050522.exp    : experience data trained only with Eman
ProteusSF_030622.exp : experience data merged from Sarona's engine and Igor's data

Opening list : DeeDs 50 openings list #2

Code: Select all

   # PLAYER                              :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_050522.exp (Eman 8.10)       :      61     53    58.0     100  58.0   27   62   11  62.0       0     1
   2 ProteusSF_030622.exp (Eman 8.10)    :       0   ----    42.0     100  42.0   11   62   27  62.0      61     1

White advantage = -96.04 +/- 27.09
Draw rate (equal opponents) = 50.00 % +/- 0.00

link : games
pass : ilUjtD1BLyiiL3gGfA9ihA

[deeds]

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

Code: Select all

Depth4_050522.exp    : experience data trained only with Eman
ProteusSF_030622.exp : experience data merged from Sarona's engine and Igor's data

Opening list : Jeroen Noomen's 50 tcec season 22 superfinal openings

Code: Select all

   # PLAYER                              :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_050522.exp (Eman 8.10)       :      85     50    61.5     100  61.5   24   75    1  75.0       0     1
   2 ProteusSF_030622.exp (Eman 8.10)    :       0   ----    38.5     100  38.5    1   75   24  75.0      85     1

White advantage = 71.44 +/- 25.88
Draw rate (equal opponents) = 50.00 % +/- 0.00

link : games
pass : ilUjtD1BLyiiL3gGfA9ihA

[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)

[deeds]

So by "merged" experience data you mean data from different engines right?

Yes as explained here (by me) and confirmed there (by Sarona).

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 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

[EmptikBest]

So by "merged" experience data you mean data from different engines right?

Yes as explained here (by me) and confirmed there (by Sarona).

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 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