NNUE - only from own engine?

[AndrewGrant]

I support TCECs rules.

But like all good intentions, it will hurt some of the honest people, and reward the dishonest. I find reusing the code of Stockfish and Leela, their games, their tools, their exact implementations, their methods, to be boring. Do something different, or join the people doing the same thing.

[Rebel]

I believe most of the major rating list testers (CEGT, CCRL and others) are not interested in SF NNUE unless the net is trained solely on the engine's own games

That's true.

I think the issue is important enough open a discussion.

NNUE eval is the result from :

1. Training software (freely available)
2. Quality of the EPD's (hard own work)
3. NNUE implementation (freely available)

Since the elo is in the EPD (and is the gold digging part) I see no good reason reason to put a limitation on the creative part (the quality of the EPD's). Further it is discrimination to starters who are forced to write a good HCE eval first. Everybody should be free to create his own EPD database as he pleases, train it as he pleases, implement it as he pleases.

I see only one limitation, using an existing NNUE from someone else. It means the creative (and hard) part is skipped. Fire comes to mind, I don't test it.

Lately, nearly all of the progress for SF nets have been made in the trainer - not the data. What you call the 'EPD' part (besides the only EPDs used are the opening books for data generation, EPD itself is not suitable for training)

I believe it's possible to train from an EPD database, if each position includes an evaluation.

Is this what Ed is suggesting?

Has anyone tried this?

I have argued in the OP for a minimum requirement that authors at least should create their own NNUE.

[JohnWoe]

Stop the press !
Somebody is using open source components in their open source program !

[amanjpro]

Stop the press !
Somebody is using open source components in their open source program !

Data is not open source, data is not code.... Not saying one cannot use it (or should not use it).. But calling it open source is just misleading

In this particular discussion it is about making a different engine that feels more or less different that is worth testing and spending time on.

If I bring SF net, use it as is that is boring
If I bring readily available data, and use it in a novel way that is more interesting

[matejst]

From past experience, experimentation, innovation leads to new breakthroughs. Engineering and optimisation do not. Trying something different is the way to improve, while copying leads to ultimate failure. That's the reason why originality is so important in the long run.

Let's just take the example of NNs. Matthew Lai implemented it in Giraffe. Peter Osterlund showed that it was possible to use them for evaluation on a CPU. Jonathan Rosenthal trained several nets in 2019 already. All these experiments did not bare fruits immediately, but were used in Go, and landed in SF. I have no doubts that another great ideas will be implemented, tested and showed as valid, and then optimised in one of the big team projects.

[ChickenLogic]

Stop the press !
Somebody is using open source components in their open source program !

Data is not open source, data is not code.... Not saying one cannot use it (or should not use it).. But calling it open source is just misleading

In this particular discussion it is about making a different engine that feels more or less different that is worth testing and spending time on.

If I bring SF net, use it as is that is boring
If I bring readily available data, and use it in a novel way that is more interesting

And since when can only code be free and open source? Leela's and SF's data sets are all free as in libre and open as in open in telling one exactly how it was created and how to reproduce it. I could easily produce a data set that you have to
a) buy (not that open source can't be monetized)
b) use in a certain way.
I certainly won't tell you anything on how to create your own spin of it either.

I do agree with the rest of the statement. Use the data you want.

[amanjpro]

Stop the press !
Somebody is using open source components in their open source program !

Data is not open source, data is not code.... Not saying one cannot use it (or should not use it).. But calling it open source is just misleading

In this particular discussion it is about making a different engine that feels more or less different that is worth testing and spending time on.

If I bring SF net, use it as is that is boring
If I bring readily available data, and use it in a novel way that is more interesting

And since when can only code be free and open source? Leela's and SF's data sets are all free as in libre and open as in open in telling one exactly how it was created and how to reproduce it. I could easily produce a data set that you have to
a) buy (not that open source can't be monetized)
b) use in a certain way.
I certainly won't tell you anything on how to create your own spin of it either.

I do agree with the rest of the statement. Use the data you want.

Data doesn't have a source code to be open source, it can be publicly available... But that is technicality

[ChickenLogic]

Then call it free and open data.