FCP Tourney-KI started January 02nd, 2022

[Graham Banks]

Just keep doing what you enjoy doing, Frank.

It's not always a good idea to let the views of others hurt you.
Being a moderator for three terms in Talkchess taught me that.

[Frank Quisinsky]

Hi Graham,

yes, same experience in all the chess fora I am admin in the past.
Conner is important for me, I do that for him.

I remember me on your time as admin.
Different situations are very complicated and a lot of comments against your person were not fair.

You are right!

Best
Frank

[kranium]

Enough ...
For the moment I will changed nothing with name of Fire neural network.

Frank, if you feel the need to rename it, I would suggest NN-SV

SV standing for Sergio Vieri, the Indonesian computer scientist who created it and trained it:
https://www.chessprogramming.org/Sergio_Vieri

Unfortunately he no longer participates in the development of state-of-the-art reinforcement and supervised learning techniques
with respect to computer chess.

It is not comparable to a copy and pasted SF eval function, it plays uniquely...don't let yourself be mislead in that regard.

Best Regards,
Norm

[Rebel]

Enough ...
For the moment I will changed nothing with name of Fire neural network.

Frank, if you feel the need to rename it, I would suggest NN-SV

SV standing for Sergio Vieri, the Indonesian computer scientist who created it and trained it:
https://www.chessprogramming.org/Sergio_Vieri

Unfortunately he no longer participates in the development of state-of-the-art reinforcement and supervised learning techniques
with respect to computer chess.

It is not comparable to a copy and pasted SF eval function, it plays uniquely...don't let yourself be mislead in that regard.

Best Regards,
Norm

Just create your own net like everybody else. I know it is a pain, but without pain no glory.

[Frank Quisinsky]

Hi Norman, Ed,

for Norman:
If I do that tomorrow the next comment ...
Please rename Dragon 3 NN (Komodo) to Dragon 3 NNDK (Komodo).

I need a better solution.

NNSf is also not right.
NNLC (Leela Chess) is more right.

But if i used NNLC what is with all the others NNSf programs in my list.
I have to rename different of these programs in NNLC too.

Again ...
I need a better solution.
I need many hours to rename an engine in my list because I have to do that for all the statistic files, databases etc. too.

---

For Ed:
A net is for sure a lot of work and an important part of a chess program.
A "part" of a chess program ... not more and not less.
Without a net a chess program works fine, we all know that.



With a net only we can nothing do.
Not a glory ... an important part yes but for a glory others things are much more important.

For the moment I changed nothing.
Furthermore, my tourney-system is not very important.
More CCRL or CEGT and the others like FGRL, SPCC, Ipman ...

I collect material only for an other project I have in my brain to do and will give John Stanback material.
Own material in a development is better as to use material by others.
The Wasp programmer have an own net with own games and on Wasp site an own rating-system.
And the Wasp programmer since Zarkov, a lot of own ideas, working on GnuChess ... same Fabien do a time and gave never the information to Stefan Meyer-Kahlen ... your engine based on GnuChess. Often heard that MChess based on GnuChess, no idea if right. Comet based on GnuChess. A lot of originality ...

After run-8 I hold my tourney-system private available.
A reason more ... my material is not very important!

Best
Frank

[Frank Quisinsky]

Hi Ed,

your posting to the message from Pedro ...

Pedro made a good point, you too ...
Can be use for ...

NNv1 = own net based on games with own engine
NNv2 = own net based on games with own and other engines
NNv3 = own net based on games from other engines
NNv4 = own net based on games comes from humans
NNv5 = own net based on a mix of humans and engine games
--
NNv6 = net based on an other engine with own material (the programmer have to give an explanation on his github or web site).
NNv7 = net based on an other engine (the programmer have to give an explanation on his github or web site).

NNv = Neural Network variant ... x

Programmers can give the information, collect in a new thread in TalkChess in front of all others.
And the problem is solved, not for my work only.

Such things are interesting, not what I do in my list.

Confused, a mess ...
Same in the time all programmers are working on own GUIs without to support engine protocols.

Best
Frank

[Modern Times]

Just create your own net like everybody else. I know it is a pain, but without pain no glory.

Well, In the business world, you don't reinvent the wheel. That costs time and money. If something is there, publicly available and free to use, and any licensing attached to it enables you to use it, then that is what you do unless there is very good reason not to. That is the sensible and logical approach. Why is chess any different ? And who says it has to be different ? And who decides ? And who is the judge ? There are programmers who want to do everything themselves for the challenge and personal satisfaction, and I applaud that. And then there are others who choose to save time - in Norm's case above, plugging in a net developed by Sergio Vieri - which then enabled him to focus his time on other areas. Prioritising your time and effort is also common practice in both business and personal lives. No-one has unlimited time.

So as HG and others have said, at the end of the day it is up to individual tournament organisers to decide for themselves what to include and what not to include. If someone does not like what a particular tournament chooses to run, or how they name the engines, then they aren't forced to look at it. Plenty of options available out there.

In terms of nets, has dkappe not developed nets for other engine authors, or at the very least collaborated heavily with them ? Where do you draw the line there ?

[Frank Quisinsky]

Modern Times,

thank you!!!
In much better english as my frankish.

100% what I am thinking!

Best
Frank

[kranium]

...as Fire uses an old SF network taken directly from FishTest which was trained on SF evaluations.
Note that this amounts to copy pasting Stockfish's evaluation function into one's own engine.

Nonsense. Fire-NN plays uniquely, as evidenced by the Sim data when compared to the last 4 versions of SF:
http://users.telenet.be/chesslogik/images/matrix.jpg
40-43%, as if the 2 engines are completely unrelated.
Any 3400 Elo engine that plays 50-60% different moves than SF and Komodo is interesting and valuable.

That's a remarkable thing about NNUE, even if 2 NNUEs are trained from the same or similar data, they can (and generally do) vary significantly.

Just create your own net like everybody else. I know it is a pain, but without pain no glory.

Not interested in kowtowing to you...

PS It's NOT everybody else, as I can probably name 6-8 engines (and more are appearing on github weekly) that are using data (fens/pgns) from other engines (not to mention Stockfish, and LC0).
This includes Mayhem and Coiled. Unfortunately, these significant and valid efforts are being ignored by some individuals & testing groups because individuals like you and Connor believe they can dictate the rules...and often it crosses the line to bullying.

[connor_mcmonigle]

...as Fire uses an old SF network taken directly from FishTest which was trained on SF evaluations.
Note that this amounts to copy pasting Stockfish's evaluation function into one's own engine.

Nonsense. Fire-NN plays uniquely, as evidenced by the Sim data when compared to the last 4 versions of SF:
http://users.telenet.be/chesslogik/images/matrix.jpg
40-43%, as if the 2 engines are completely unrelated.
Any 3400 Elo engine that plays 50-60% different moves than SF and Komodo is interesting and valuable.

That's a remarkable thing about NNUE, even if 2 NNUEs are trained from the same or similar data, they can (and generally do) vary significantly.

Just create your own net like everybody else. I know it is a pain, but without pain no glory.

Not interested in kowtowing to you...

PS It's NOT everybody else, as I can probably name 6-8 engines (and more are appearing on github weekly) that are using data (fens/pgns) from other engines (not to mention Stockfish, and LC0).
This includes Mayhem and Coiled. Unfortunately, these significant and valid efforts are being ignored by some individuals & testing groups because individuals like you and Connor believe they can dictate the rules...and often it crosses the line to bullying.

I'm not dictating the rules. I'm attempting to correct misinformation and others' misconceptions. Frank seemed to think you had produced your own network which is definitively false. Engines like Fire, Mayhem and Coiled all use inference code derived from SF and networks taken directly from the Stockfish project which were trained using SF data, training tools and architectures. These engines all have zero original effort in their evaluation functions. Note that many of these engines' authors recognize using Stockfish networks as a stop gap measure until they can produce something of their own, being early in development.

Perhaps unsurprisingly, such engines' evaluations are consequently highly correlated (in the statistical sense) with Stockfish's evaluations relative to engines which don't copy previous versions of Stockfish's evaluation function verbatim. I'll share some data demonstrating this phenomenon later today.