Andscacs nnue 0.1

[Frank Quisinsky]

Hi there,

Let us collect a motivation packet for the programmer of Andscacs!
Much more interesting as to start a bad discussion!

March 2014 ... the IPP problem!
TOP-20, FCP Rating List, final ...

Code: Select all

   1 Houdini 4 STD x64 A              : 3128.3    1142.5    1450   78.8%
   2 Stockfish 08.02.14 SSE42 x64     : 3105.6     989.5    1250   79.2%
   3 Stockfish 05.12.13 SSE42 x64     : 3097.5     994.5    1350   73.7%
   4 Houdini 3 Pro x64                : 3087.4    1035.0    1390   74.5%
   5 Komodo TCECr x64                 : 3073.5    1159.0    1600   72.4%
   6 Stockfish 03.11.13 SSE42 x64     : 3066.1     727.5    1000   72.8%
   7 Houdini 2.0c x64                 : 3054.7    1111.0    1400   79.4%
   8 Komodo 6 x64                     : 3044.8     814.5    1170   69.6%
   9 GullChess 2.8 Beta BMI2 x64      : 3038.6     927.5    1300   71.3%
  10 Houdini 1.5 x64                  : 3031.4    1814.5    2320   78.2%
  11 Critter 1.6a x64                 : 3015.8    1017.5    1540   66.1%
  12 Critter 1.4 x64                  : 3012.5    1244.5    1600   77.8%
  13 Houdini 1.5 w32                  : 3008.9     976.5    1240   78.8%
  14 Equinox 2.02 x64                 : 2997.3     893.0    1400   63.8%
  15 Equinox 2.00 x64                 : 2997.1     869.0    1340   64.9%
  16 Komodo 3.0 x64                   : 2990.3    1203.0    1600   75.2%
  17 Fire 3.0 AVX x64                 : 2989.1     740.5    1100   67.3%
  18 Bouquet 1.8 x64                  : 2989.0     896.0    1440   62.2%
  19 Komodo 4.0 x64                   : 2988.8    1208.5    1600   75.5%
  20 Rybka 4.1 Exp.79 x64             : 2987.1     942.0    1200   78.5%
  

Thinking on times before Shredder is a long time on place 1.
All the programmers in the past must be very stupid if programs started today with 2800 Elo or have over 3000 Elo. Shredder was a long time, with around 100 Elo more as the others, on place 1 with 2800 Elo.

The great new Cheng 4.40 vs. Shredder 12 (World Champion, long years on place 1 with 2800 Elo).
Vs. Cheng 4.40 no chance for Shredder 12!

Are today all the programmers "Dirty Nasty Cloners"?

No, no ... intelligent people!
Searching all the time for the next good ideas in chess programming!
The causal chain!

NNUE is for a programmer a very boring work.
To train on own net ... sure, most do not like it.

Programmers should work on new ideas and not waste on time with old ideas.
NNUE is today an old idea but maybe during the work, NNUE support add, talents in chess programming can create in additional new ideas. A good job for programmers, like Daniel and many others.

Try it out and thinking about it ...
Sure, the programmer of Andscacs do that.

The year 2014 ... the IPP year.
Since a while we celebrate the ... Stockfish years.
Since a while we celebrate the ... Shogi ideas.

How many of TOP-50 chess programs used good known old ideas?
95% or 99% or 100% ... please help me, not sure here?

The programmer of Andscacs made some experiments with NNUE and give us the work and his impressions.
What is wrong, nothing is wrong!

More originality = an own net

The NNUE idea is always the same, end of the year 2020 available with small changes by a group of around 15 programmers. It's good if programmers try out NNUE and are able to add own ideas. But in reality ... the NNUE idea is the same.

To give a small group of "exclusive programmers" the OK ...
You are an TalkChess supporter, here you have, you can use NNUE.

And for the others no, no ... you are a "Cloner" ... that's a typical "HNP" ... human narrow perspective.

What people like to do if the day is long and the night is short.
Not new for me!

Best
Frank

[Frank Quisinsky]

There are some true heroes of computer chess.
Claude Shannon
Robert Hyatt
Ed Schroder
Jon Dart
Fabien Letouzey
Hsu/Campbell
Tord Romstad
Ronald de Man
Bojung Guo

For me ...
John Stanback ... the grandvather of chess programming, working around 40 years on it.
No other from the "older" programmers develops such a strong chess program with so many own ideas and energy.
The maestro of motivation and energy, I celebrate each mail from John Stanback and I am happy that I have contacts to him.
100% motivation packet for myself.

But the list is fully Ok for me, my number 1 missed only.
No problem to give in additional the missed name.



Best
Frank

Ah, my number two (Stefan Meyer-Kahlen) missed also.
A lot of people missed here ... but my number three and four (Fabien and Tord are inside).
That's good!
Robert Hyatt is really a Prof.
Good to have Bob and I can fully understand that Robert Hyatt is for many others the number 1.
But for myself I have a clear number 1 after 44 years computer chess.

[Frank Quisinsky]

Texel programmer missed.
Don Daily missed ...

Insomniac programmer missed, with 14 years on Crafty level, long years ago ... we lost him.

Frans Morsch after a long discussion in Leiden about chess programming with James Robertson ...
I asked him ... Frans told me "What for a talented programmer".

So many others ...

[supersharp77]

There are some true heroes of computer chess.
Claude Shannon
Robert Hyatt
Ed Schroder
Jon Dart
Fabien Letouzey
Hsu/Campbell
Tord Romstad
Ronald de Man
Bojung Guo

For me ...
John Stanback ... the grandvather of chess programming, working around 40 years on it.
No other from the "older" programmers develops such a strong chess program with so many own ideas and energy.
The maestro of motivation and energy, I celebrate each mail from John Stanback and I am happy that I have contacts to him.
100% motivation packet for myself.

But the list is fully Ok for me, my number 1 missed only.
No problem to give in additional the missed name.



Best
Frank

Ah, my number two (Stefan Meyer-Kahlen) missed also.
A lot of people missed here ... but my number three and four (Fabien and Tord are inside).
That's good!
Robert Hyatt is really a Prof.
Good to have Bob and I can fully understand that Robert Hyatt is for many others the number 1.
But for myself I have a clear number 1 after 44 years computer chess.

Tom Kerrigan (TCSP)
Martin Bryant Colossus Chess
Mr Hiarcs
Mr Shredder..Stefan Meyer-Kahlen
Mr Deep Junior...Amir Ban and Shay Bushinsky
Deep Blue...Feng-hsiung Hsu
Chess System Tal....Chris Whittington
Chessmaster....David Kittinger..Kathe Spracklen..Dan Spracklen...Johan de Koning
Pawel Koziol...Rodent and Open Tal
Dieter Bürßner....Yace Chess
Lex Loep....LChess (Lokasoft)
Matthew Lai....Giraffe
Vincent Diepeveen,...Diep Chess
Peter Österlund...Cuckoo Chess..Texel
Alberto Alonso Ruibal...Carballo Chess (Java)
Krasimir Topchiyski...Bagatur Chess Java
Gian-Carlo Pascutto..Sjeng Chess & LC0...

Well..that's a start....AR

[Frank Quisinsky]

Hi there,

not Vincent Diepeveen after Leiden (gave the information Ruffian is a clone from Shredder, and after Theo checked the sources ... no sorry).
not Chris Whittington ... can nothing see what can be interesting, Chess System Tal plays more boring chess as Chess Tiger. A myth build since years by a handfull persons, not more not less. The programmer should give the missed proof. I have many games and a very bad "short-game" statistics. Same for ChessTiger and Gambit-Tiger ... a myth, not more no less. Same for Rybka, most boring chess in mid-games with a lot of mistakes.

But all the others, yes ... means don't know Alberto Alonso.
A good job for the evening for search information about him.



Best
Frank

Bagatur ... also a very nice software, without any doubt.
Feng shui?
Ah Feng Hsu ... a standard chess program with a good SMP support (30 others are stronger at this time) on a strong hardware.
Not a VIP in computer chess for me.

[Frank Quisinsky]

Hi,

I am sure Daniel José Queraltó will read the thread ...

Daniel, which engines you like most?
I am sure we can learn a bit if you told us your opinion?

Or better ...
From which programs and ideas you learn most?

Best
Frank

[mar]

Now, of course, I think people should write their own code, or use code with permission.
But it seems like we don't even want people to learn from each other.

I think you possibly misunderstand what this is. The author has effectively copy-pasted Stockfish's evaluation function into their engine (the evaluation functions match exactly). This could be an interesting experiment to see how an engine's search compares to Stockfish's search, but hopefully it's not intended to be anything more. If rating lists bother to test such engines using Stockfish's evaluation function, they will quickly be over run with partial Stockfish clones such as this.

I too hope that people will do interesting things with neural networks. I am firmly of the opinion that there exists at least as much room for originality when using neural networks as evaluation functions as exists when using linear models as evaluation functions. With a linear model, you solely have control over the input features. With a neural network, you have both freedom in input features and architecture. It's been a long time since a top level engine featured a hand tuned evaluation function so the argument that using an optimization algorithm to adjust the many parameters found in modern evaluation functions leads to something unoriginal is nonsense...

Another disappointing aspect of this release is that it's in blatant violation of the GPL. Andscacs, as a closed source engine, cannot rely on the GPL'd NNUE code it copies unless the entirety of the source code is also distributed to the users along with the binaries...

funny that this comes from someone who used SF to label the training set. to each his own.

[willmorton]

I'll stick to plain old Andscacs. There are more than enough Stockfish nets floating around.

I don't know about the others but your constantly picking on other programmers as soon as they publish a NNUE version of their engines, asking if they are using an original net or if they have used Stockfish to train their nets and son on, let alone the sad pitying smiles because now the engine is no longer original, is becoming annoying to me. You made your point many times, there's no need to popup each time a NNUE engine is released always repeating the same stuff.

Daniel has released a NNUE version of Andscacs just for fun and showed how much stronger it becomes using a very strong Stockfish net. If and when he wants, he'll release his own net trained with whatever he pleases. People running rating lists will decide whether to test it or not. They don't need you to constantly suggest them which engines to include and which ones to remove.

Asking people not to use Stockfish/Lc0 or anything else for tuning is simply silly especially after you used yourself Stockfish labeled positions in the past.
If you are right, Deep Blue should have not used games from the strongest chess entities (human GMs at that time) for automatic tuning, human players should not study the books/moves of best players but only their own games, no engine should use opening books based on the strongest engines games,...

And by the way, it is fun that if someone makes some minor changes to the NNUE arch (eg removing the simmetry) suddenly he has his "own" original net, while the NNUE key points (KP and incremental updates) are all still there...

[Daniel Anulliero]

I'm all right with wilmorton
Daniel said since in his first post , he did Andscacs NNUE just for fun and curiosity

"I was curious about what could I obtain adding nnue to Andscacs."
and :
"Of course rating lists that care about engine originality should not add this version."

he didn't violated anything

It is not necessary to be too bad with him

Chess computer is taking a strange way now , very pity

I'll try one day to add sf nnue to Isa , sure I'll win 600 elo , as my eval is sure very bad

[AndrewGrant]

I'll stick to plain old Andscacs. There are more than enough Stockfish nets floating around.

I don't know about the others but your constantly picking on other programmers as soon as they publish a NNUE version of their engines, asking if they are using an original net or if they have used Stockfish to train their nets and son on, let alone the sad pitying smiles because now the engine is no longer original, is becoming annoying to me. You made your point many times, there's no need to popup each time a NNUE engine is released always repeating the same stuff.

Well there is a reason actually. Author clearly stated that rating lists should not pick this up, as its his experiment out of curiosity, not a serious release intended to be the new face of his engine. This has happened with most all other NNUE releases. From Nemo to Rubi to Igel, whenever they used Stockfish nets, the authors were clear that these were just for curiosity [MYSELF INCLUDED] But rating lists rate them anyway.

Now you have 8+ Stockfish versions on every single rating list. That is annoying (obviously, especially Fire and Houdini, who lied), but also distorts the meaning of those ratings as a pool with 75%+ of engines having high similarities makes the sample no longer representative of chess engines as a whole.

You are free to place me on ignore if you don't like or understand my comments. As are all others. I have the freedom to type, and you have the freedom to not be forced to read the words I type. If you block me, I wish you the best of luck. Cheers.