Andscacs 0.1 with NN file from Coiled ...

[kranium]

Apologize for the length, I don't post much these days.

Just to clarify:
Fire NN search has nothing to do with SF search. It's entirely unique.
It achieves a much wider search, generally getting many more nodes examined, but on average 8-12 ply less deep.
All this can be evidenced: Fire-NN is open-source and available for inspection on Github.

Fire does use an NNUE eval that was created by processing Stockfish games. It was created as a candidate for SF release, but it failed testing,
and was never used. It's not owned by SF, and is pure data, public domain. It's a concentrated repository of positional evaluation knowledge.
Why shouldn't programmers be able to use it? That's the meaning of open source.

Fact is, even when using an SF NN, engines play uniquely, as there are so many other parts to the equation...
normally playing the same move only 50-60% of the time. And this is for top 3500 elo engines!

Zero NNUE code from SF is used in Fire..
The NNUE interface code is a unique open-source publication from Daniel Shawul, and I don't know of only 1 other engine that uses it.
The effort of implementing stable tuned NNUE and MCTS took me 2-3 months of intense programming and testing.


Secondly, the engines mentioned above:
Gogobello, Mayhem, Tornado, AndscacsNNUE, Rodent NN, and quite a few others, are ALL valid engines! and are likely the result of some very real effort and hard work by real people. They all deserve to be tested and evaluated!

We should be encouraging open source development, not summarily dismissing it. I realize the poor testers/groups have to decide...and they are pushed, criticized, threatened with boycott, etc., and influenced by hate and misleading innuendo and misinformation. Test anything reasonable then instead of being judge and jury.

There's far too much hate and bias, and the singular overriding motivation by many is to pump themselves up at the expense by discrediting others.

How can this forum and it's unskilled 'judges' so easily dismiss that effort and simply toss that work/effort away?
Why aren't testers, and other members here in the business of promoting and encouraging all efforts of open-source development?
Instead, we see only a parade of criticism and efforts to discredit other peoples efforts as 'lazy', or unoriginal, etc. when in fact the individuals doing this work are doing something extremely difficult and time-consuming. What do they receive in return?

There are 'programmers' here who call using anything related to SF NNUE 'copy-paste'...
Sadly, these are likely the ones who are: unable to implement NNUE in their own engines, having difficulty, or don't understand it well enough.
They seem to find solace in criticizing and downplaying the effort of others.

Additionally, there are programmers here who have taken a very different 'unique' direction with NNUE, unfortunately, they too participate regularly with negative innuendo and dismissive rhetoric about any engine that has not followed suit.

It hurts to see programmers and others here participating in a judgmental mud-throwing contest, instead of enthousiastic encouragement.

Oh God help us, what a world we live in...
Ok I've vented enough...

I wish Peace for Ukraine and the World!

[connor_mcmonigle]

..
Fire does use an NNUE eval that was created by processing Stockfish games. It was created as a candidate for SF release, but it failed testing,
and was never used. It's not owned by SF, and is pure data, public domain. It's a concentrated repository of positional evaluation knowledge.
Why shouldn't programmers be able to use it? That's the meaning of open source.
...

It's a network trained on Stockfish games by Stockfish contributors with the intent of using it with Stockfish. I think it's fair to call it a Stockfish network. This network originates from the same RL run that yielded the network used with SF 12 and evaluates positions quite similarly in my testing.

...
Fact is, even when using an SF NN, engines play uniquely, as there are so many other parts to the equation...
normally playing the same move only 50-60% of the time. And this is for top 3500 elo engines!
...

With the same network loaded in SF, it plays quite similarly and evaluates positions almost identically to Stockfish as one would expect given they're using identical evaluation functions.

...
Zero NNUE code from SF is used in Fire..
The NNUE interface code is a unique open-source publication from Daniel Shawul, and I don't know of only 1 other engine that uses it.
The effort of implementing stable tuned NNUE and MCTS took me 2-3 months of intense programming and testing.
...

Daniel Shawul's NNUE probe code is taken from CFish and is functionally identical to that which was found in SF. All the aforementioned engines (Gogobello, Mayhem, Tornado, AndscacsNNUE, Rodent NN), use the exact same probing code. With Daniel Shawul's probing code, bolting SF's evaluation function on to an existing engine should take a couple of hours at most.

...
Secondly, the engines mentioned above:
Gogobello, Mayhem, Tornado, AndscacsNNUE, Rodent NN, and quite a few others, are ALL valid engines! and are likely the result of some very real effort and hard work by real people. They all deserve to be tested and evaluated!
...

It's understandable that the majority of testers are uninterested in testing a bunch of engines with identical evaluation functions. Testers are under no obligations to test such engines or any engine for that matter. Copying SF's evaluation function functionally verbatim is comparatively low effort so it's a bit disappointing to see these engines displace more original efforts (such as Cheng for example) in terms of time allocated to testing engines. Otherwise, I've no qualms with testers testing such engines provided the nature of such engines is made clear in the rating list (I think Frank's aforementioned solution is quite reasonable).

[Sopel]

The less original an engine is the less interesting it is. It is natural that less interesting engines are not being tested as much.

[kranium]

Daniel Shawul's NNUE probe code is taken from CFish and is functionally identical to that which was found in SF. All the aforementioned engines (Gogobello, Mayhem, Tornado, AndscacsNNUE, Rodent NN), use the exact same probing code. With Daniel Shawul's probing code, bolting SF's evaluation function on to an existing engine should take a couple of hours at most.

You don't seem to understand? Yes that's the idea. It's open-source... working!

Everything on github is happily shared. There are no game programmers griping if someone uses their platform game tileset generator for a new game project. On the contrary, they're pleased. And the game developer has saved countless hours of work...hours now free to concentrate on improving other areas of his game.

That's what SF is all about. It's not a pissing contest.

[kranium]

The less original an engine is the less interesting it is. It is natural that less interesting engines are not being tested as much.

I can't argue with that. Individuals indeed should be completely free to choose.

But likewise, CCRL, CEGT, etc. should not be pressured by programmers here to test/not test certain engines.
Unfortunately that happens all to often.

[connor_mcmonigle]

Daniel Shawul's NNUE probe code is taken from CFish and is functionally identical to that which was found in SF. All the aforementioned engines (Gogobello, Mayhem, Tornado, AndscacsNNUE, Rodent NN), use the exact same probing code. With Daniel Shawul's probing code, bolting SF's evaluation function on to an existing engine should take a couple of hours at most.

You don't seem to understand? Yes that's the idea. It's open-source... working!

Everything on github is happily shared. There are no game programmers griping if someone uses their platform game tileset generator for a new game project. On the contrary, they're pleased. And the game developer has saved countless hours of work...hours now free to concentrate on improving other areas of his game.

That's what SF is all about. It's not a pissing contest.

You're free to copy Stockfish verbatim besides for altering the name if you so please. However, effectively no one will have any interest in your project if you do so unless you are successful in misleading them. What you interpret as bullying testers largely consists of attempts at correcting testers' misconceptions/offering opinions to testers.

The text you've quoted above was in response to your claim that Fire is the only engine to use Daniel Shawul's probing code ("... I don't know of only 1 other engine that uses it."). Several other engines (enumerated in the quoted text) use the probing code. You also claimed that " Zero NNUE code from SF is used in Fire.." which is rather misleading as the probing code you're using is from CFish which is in turn derived from the implementation in SF.

[Frank Quisinsky]

Hi there,

in my opinion all the programs used SF NN or a variant of SF NN (based on Stockfish: Whatever it is in detail) must be test. With the NN file alone a chess program not lost the own face, playing style of program is the same. Can be see in stats very easy.

If CEGT, CCRL or others not test this group of program's it is absolutely wrong.
But all the work around the rating systems comes from private poeple and it is the decision by the people where are working on it.

In my opinion:
Never we should give the information:
Programm x is a better work as programm y.
Each program is a good work and welcome for the community.

What we need is a much better licence model for NN files, every programmer can use without any problems. The ideas around NN comes not from Stockfish.

To use available NN files at first and to develop own NN ideas as second is a good idea and we should support that.

Very good known that all the years again and again people like to play the policemen or god.
That's the biggest problem in computer chess fora all the years again and again.

Mister X will be a good friend from Mister Y.
Mister Z don't like Mister A.
For Mister B the number 1 is "god" and all around it is right.
For Mister BA the number 2 is "god" and the number "1" is the devil.
All what have Mister F in the brain is right, what all others are thinking is bad.
If Mister A, B, C have a very clear opinion in combination Mister D have the same.

1000 different situation ...
We have to learn to accept.

I await as customer of a chess program a new version as first. I am customer and give my confidence!
If people working on rating systems will get such commercial material at first it is "psychological influence" on a very high level.

Case of Ethereal!
Such things I dislike much more as a harmless programmer used an NN Stockfish file.

Best
Frank

[Frank Quisinsky]

If a programmer (example the Minic programmer or the programmer of Wasp) working months, years ... in case of Wasp ... John is working 18 months on own NN ideas ... I should give a hint to others.

Means, if I have - as private person- interest to public ratings on a website ...
others can find it must be clear ...
that this important part of program is an completly own development.

Here I made different errors because I don't have to 100% knowledge about it.
I think the solution I found is OK, not the best way but OK.

Here I have all the time headache!

[Frank Quisinsky]

Hi there,

forgot: Ethereal 13.50 was an example:
Nothing I like more as to add Ethereal 13.50 in my tourneys.

A big conflict with myself!
No way after the last very short email disucssion with Andrew.

To accept and to try to understand others is more interesting as to search the confrontation.
Maybe I will find in the future a way for a new disucssion with Andrew (by mail).
I am open for such things!

Often I believe ...
I like the work Norman do so many time ...
Maybe for others the reason to search the war?

With group buildings in computer chess never we are able to search the more friendly way!!
It should be possible to start an open discussion about a conflict situation ... more friendly!

Best
Frank

[Frank Quisinsky]

Ah I forgot ...

And here the file the programmer of Andscacs added in a TalkChess message for around 1 year (January 2021).

Same so many others do ...
How strong can be the program with the NN file from Stockfish was the question for fans from Andscacs!
Without a test we can give not an answer!
Such orders are orders for rating systems!

Same the programmer from Greko do
Same the programmer from Rodent do
Same the programmer from Gogobello do
Same the programmer from Caissa do
Same the programmer from Orion do
Same the programmer from Coiled do

And different others if I read the thread here!
In my opinion nothing is wrong!

This executable from Andscacs 0.1 will be run with older NN files.
Please rename the "nn-file" in ...
nn.bin

https://www.amateurschach.de/download/andscacs_v0.1.zip (without nn file).

Maybe others can give information with NN file can be interesting for Andscacs 0.1.

Best
Frank