opening book vs experience file

[deeds]

Why you do not share your exp file ?
Where other chess friends may benefit...

Be generous at least once...

Your ignorance is unbearable !

I've shared more experience files than all engine trainers put together :
31.05.2021
03.06.2021
09.06.2021

I even got banned so other chess friends could benefit :
11.10.2021

[deeds]

So don't mess with me about "being generous at least once", I've already been generous tens of times :
tools
solutions

You organize tournaments with the same cheaters who motivated all engine trainers to remain in the private sphere !

[Sedat Canbaz]

Why you do not share your exp file ?
Where other chess friends may benefit...

Be generous at least once...

Your ignorance is unbearable !

I've shared more experience files than all engine trainers put together :
31.05.2021
03.06.2021
09.06.2021

I even got banned so other chess friends could benefit :
11.10.2021

Oh..really sorry for saying
'be generous at least once...'

But anyhow, actually my real question was regarding
Your latest exp file, which is under test (vs HCanFree book)

Do you plan to share it..?

Thanks in advance

[cubail]

DRAGON-7, Blitz 2.0min+1.0sec 4 core frıtz17 gui syzgyz 6 man Intel(R) Xeon(R) CPU E5-2690 0 @ 2.90GHz play 200 games

1 2 3 4 5
1 AI 27.0_ad9b x64m ** 0.0/0 0.00 ---5.8 mb expencıve file play
2 SpecTral GNX 2.40_m ** 0.0/0 0.00 ---6.6 mb expencıve file play
3 RapTora 2.2-modern ** 0.0/0 0.00 ---5.9 mb expencıve file play
4 EMAN 8.91 64-bit SSE41,POPCNT ** 0.0/0 0.00 ---1.5 gb expencıve file play
5 EMAN 8.91 64-bit SSE41,POPCNTa ** 0.0/0 0.00 ---HCanFree-Upd.bin
test start

DRAGON-7, Blitz 2.0min+1.0sec 2023

1 2 3 4 5
1 RapTora 2.2-modern 0.5 - 0.5** 0.5/1 0.25
2 AI 27.0_ad9b x64m 0.5 - 0.5 ** 0.5/1 0.25
3 EMAN 8.91 64-bit SSE41,POPCNT 0.5 - 0.5 ** 0.5/1 0.25
4 EMAN 8.91 64-bit SSE41,POPCNTa 0.5 - 0.5 ** 0.5/1 0.25
5 SpecTral GNX 2.40_m ** 0.0/0

[Anton101]

Nice!

Many thanks!

[deeds]

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

Opening list : 20 first legal moves

Code: Select all

   # PLAYER                           :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppAvg  OppN
   1 Depth4_190623.exp (Eman 8.40)    :      18     81    21.0      40  52.5    2   38    0  95.0       0     1
   2 hcanfree-upd.bin (Eman 8.40)     :       0   ----    19.0      40  47.5    0   38    2  95.0      18     1

White advantage = -17.58 +/- 41.39
Draw rate (equal opponents) = 50.00 % +/- 0.00

link : games
pass : 59Ef9tOGT1RCBGYgpMTuzw

[deeds]

Until Eman 8.40, the experience data contained "pawn-value" based scores.
Since Eman 8.50, the experience data contains "WDL-model" based scores.

With an experience file which contains mixed scores, it's like a merged experience file from foreign engines => the hash of the engine is polluted => the strength of the engine is reduced => unfair conditions on tourney !

The "pawn-value" based scores are higher than "WDL-model" based scores => the learning thresholds are no longer respected => wrong training settings !

All engine trainers can confirm this.

[cubail]

DRAGON-7, Blitz 2.0min+1.0sec 4 core frıtz17 gui syzgyz 6 man Intel(R) Xeon(R) CPU E5-2690 0 @ 2.90GHz play 200 games

1 2 3 4 5
1 AI 27.0_ad9b x64m ** 0.0/0 0.00 ---5.8 mb expencıve file play
2 SpecTral GNX 2.40_m ** 0.0/0 0.00 ---6.6 mb expencıve file play
3 RapTora 2.2-modern ** 0.0/0 0.00 ---5.9 mb expencıve file play
4 EMAN 8.91 64-bit SSE41,POPCNT ** 0.0/0 0.00 ---1.5 gb expencıve file play
5 EMAN 8.91 64-bit SSE41,POPCNTa ** 0.0/0 0.00 ---HCanFree-Upd.bin
all games draw

DRAGON-7, Blitz 2.0min+1.0sec 2023

1 2 3 4 5
1 AI 27.0_ad9b x64m 10.0 - 10.010.0 - 10.010.0 - 10.010.0 - 10.0** 40.0/80 1600.00
2 SpecTral GNX 2.40_m 10.0 - 10.010.0 - 10.010.0 - 10.010.0 - 10.0 ** 40.0/80 1600.00
3 RapTora 2.2-modern 10.0 - 10.010.0 - 10.010.0 - 10.010.0 - 10.0 ** 40.0/80 1600.00
4 EMAN 8.91 64-bit SSE41,POPCNT 10.0 - 10.010.0 - 10.010.0 - 10.010.0 - 10.0 ** 40.0/80 1600.00
5 EMAN 8.91 64-bit SSE41,POPCNTa 10.0 - 10.010.0 - 10.010.0 - 10.010.0 - 10.0 ** 40.0/80 1600.00

games lınk
https://pixeldrain.com/u/GLp7FbDY

[deeds]

...
all games draw
...

Default settings = training mode

Training mode = played moves are at 70-75% in conformity with their experience data. Nothing forces the engines to play their most efficient moves because the engines believe they are at training. So the learning feature can avoid the most efficient moves and can try new moves even if they are worst.

In sum, you trained 40 games per engine, and not even on the same opening = you only did a little warmup !

The last tester who did the same masquerade was Stefan Pohl years ago.

Good luck explaining to us that 1.5 GB = 6 MB hihihi

[cubail]

I started a new test, I added 2 ctg and 1 thousand books to 1+1 time

DRAGON-7, Blitz 1.0min+1.0sec 2023

1 2 3 4 5 6 7 8
1 Polyfish 230607b **************** ½½½½ ½½½ ½½½ ½½½ ½½½ ½1½ 1½1 12.5/22
2 EMAN 8.91 64-bit SSE41,POPCNT ½½½½ **************** ½½½ ½½½ ½½½ ½½½ ½½½ ½½½ 11.0/22 121.75
3 RapTora 2.2-modern ½½½ ½½½ **************** ½½½½ ½½½ ½½½ ½½½ ½½½ 11.0/22 121.00
4 Polyfish 230607a ½½½ ½½½ ½½½½ **************** ½½½ ½½½ ½½½ ½½½ 11.0/22 121.00
5 EMAN 8.91 64-bit SSE41,POPCNTa ½½½ ½½½ ½½½ ½½½ **************** ½½½ ½½½½ ½½½ 11.0/22 120.75
6 Polyfish 230607c ½½½ ½½½ ½½½ ½½½ ½½½ **************** ½½½ ½½½½ 11.0/22 120.50
7 AI 27.0_ad9b x64m ½0½ ½½½ ½½½ ½½½ ½½½½ ½½½ **************** ½½½ 10.5/22
8 SpecTral GNX 2.40_m 0½0 ½½½ ½½½ ½½½ ½½½ ½½½½ ½½½ **************** 10.0/22