Deeds on Chess Engines Testing

[deeds]

Let me ask my question in other way:
Suppose an engine like clover ( not stockfish family). Can the version 6 of clover with experience file and also skilled trainer be even stronger than version 8 of clover which uses a different and new NNUE net( but without exp file).
Thanks in advance
Note: stronger here : means having more positional knowledge and better performance.

Yes it's possible because the learned moves are loaded into the engine's hash so :
- the engine remembers the winning/losing/draw lines without having to reanalyze them
- the engine knows what the most probable endgame's material so the engine will avoid/favor certain exchanges in order to keep these pieces
- etc.

[deeds]

you can find all the tools here in this forum and creating a bin book is very easy.

Good luck to use an EXP file with contains Eman's scores with SF-FB from 09/28/24 !

[Homayoun]

And my last question
Do you think experience file is more useful for the NNUE engines that are updated very late? (For example koivisto engine hasn’t been updated for more than one year)? or it can be useful even for NNUE engines which are updated very soon. ( for example we have new version every 3-4 month with lots of changes in source code including new net,evaluation,etc. )?
Thanks in advance

[Martin_1969]

Good luck to use an EXP file with contains Eman's scores with SF-FB from 09/28/24 !

Oh boy, what are YOU talking about?
BYE

[deeds]

Do you think experience file is more useful for the NNUE engines that are updated very late ?
Or it can be useful even for NNUE engines which are updated very soon ?

For both !

At each new net, Stockfish makes fewer mistakes against itself but none of its nets allows Stockfish to win all openings at each time.
There is a variability in the Stockfish strength even with the same hardware / tc / net.

At each new training, a learning engine memorize the most efficient lines (and the lines to avoid) for a given opening.
Thanks to its EXP file, a learning engine will use them regardless of the hardware / tc / nets there is less variability.

Here, you can see few experiments :
- tc odd - eman exp vs sf15
- SF15.1 at TC 60+1 vs Eman exp. at fixed depth

Here, you can see the difference when the learning engine is helped or not by a (reinforced) EXP file :
- TC 1m+1s vs TC 6m+6s (with EXP file)
- TC 1m+1s vs TC 6m+6s (without EXP file)

[deeds]

Good luck to use an EXP file with contains Eman's scores with SF-FB from 09/28/24 !

Oh boy, what are YOU talking about?
BYE

viewtopic.php?f=38&t=4173&start=50#p37871
You said you had good performance with SF-PB with a book created from Eman/C-dragon games.
But you didn't try the same engine (=SF-PB) with the EXP file created by Eman during these same games.
That's why i answered : "Good luck to use an EXP file with a non-learning engine (=SF-PB) which has different evaluation than the engine used to create the EXP file (=Eman)

[Homayoun]

Ok. Thanks for answers.
Have good times

[Homayoun]

Hi Sedat
The only thing that I admitted here was that deeds has great and maybe unique knowledge about learning features in chess engines and also he has personal intelligence (just my opinion). But I never told that everything he says is completely right. About his personal behavior including fake accounts, disrespectful behavior with others, using bad words like liars, I have serious problems with him. I can’t accept no one of his reasons to insult others. I want to mention to something too: The word “meaningless “ that I used is not accurate. I would better to say” it’s better to test an exp file with other exp files and test a book with other books”(again my opinion, may be wrong). I want to confirm : I have serious problems with deeds behavior with other members.
Best regards Sedat

[Martin_1969]

Good luck to use an EXP file with contains Eman's scores with SF-FB from 09/28/24 !

Oh boy, what are YOU talking about?
BYE

viewtopic.php?f=38&t=4173&start=50#p37871
You said you had good performance with SF-PB with a book created from Eman/C-dragon games.
But you didn't try the same engine (=SF-PB) with the EXP file created by Eman during these same games.
That's why i answered : "Good luck to use an EXP file with a non-learning engine (=SF-PB) which has different evaluation than the engine used to create the EXP file (=Eman)

Oh little boy, what are you talking about?

[Martin_1969]

Of course I also let C_ Dragon play against SF_PB with a book. But I stopped after 210 games because the result was clear: 63.1: 36.9 for SF-PB!
It is a waste of time to play more games
games are here: https://uploadnow.io/files/HcMcRHs
and bye,bye little boy