Andscacs 0.1 with NN file from Coiled ...

[Frank Quisinsky]

Hi Conner,

most of programmers, working on an own NN file, have the same opinion!
I am more happy if all are working on own neural network files.

Again, if I am reading the mails from John about it.
John is total happy if he can make Wasp with an own nn file Elo-for-Elo stronger.
Honestly, I am too as biggest fan from Wasp and my wife also.



But from my point of view ...
I should not ignore the work by others (I am not god).
With or without an own NN file.

After "run-2" I will changed the name of engines!

1. Rodent 1.0 NNsf
2. Coiled 1.1 NNsf
3. Gogobello 3.0 NNsf
4. Andscacs 0.1 NNsf (have interest to test it, maybe a small animation for Daniel Jose to create an own NN file).
5. Caissa 0.4 NNsf (have interest to test it).

The programmer of Caissa wrote here in TalkChess a lot of messages.
Do you saw how many gave him answers?
That's not very fine!

A good idea is to give this group of engines not an own rating place.
With the information: You can get an own rating place if you create an own NN file.

Example:

Code: Select all

  11. Ethereal 13.25 NN           :  3343.54   1703    66.5   666   933   104  1132.5   54.8  10.85  3207.03  11.41   57.0
  12. RubiChess 2021 NN           :  3320.84   1699    63.9   603   966   130  1086.0   56.9  10.81  3208.84  11.33   57.9
  13. Seer 2.4.0 NN               :  3283.18   1699    59.4   510   998   191  1009.0   58.7  10.24  3209.51  11.34   57.9
  14. Igel 3.0.10 NN              :  3271.95   2179    57.6   572  1365   242  1254.5   62.6   9.10  3212.25  11.79   73.9
  15. Nemorino 6.09 NN dev        :  3257.11   2179    55.7   578  1273   328  1214.5   58.4   9.04  3212.45  11.79   73.9
  16. Arasan 23.2 NN              :  3250.96   1699    55.4   425  1032   242   941.0   60.7  10.30  3210.08  11.34   57.9
  --. Rodent 1.0 NNsf             :  3246.28   1699    54.8   390  1082   227   931.0   63.7   9.38  3210.16  11.36   57.9
  --. Coiled 1.1 NNsf             :  3246.18   1120    53.5   251   697   172   599.5   62.2  12.48  3220.34  11.16   40.0
  17. Tucano 10.00 NN             :  3239.92   1120    52.7   235   711   174   590.5   63.5  12.29  3220.50  11.16   40.0
  18. Minic 3.18 NN               :  3233.33   1120    51.9   246   670   204   581.0   59.8  12.57  3220.67  11.15   40.0
 

And to programmers do that just for a test (Andscacs, Rodent) ...
Yes, this is nice for looking.
How strong the engine can be with such a good work.

Best
Frank

[Frank Quisinsky]

Martin,

at first:
Nice to read you!

The deal?
I wrote that before ... I am just mate ... 50% to 50%, helpless is more right.

The reasons for myself:

1.
If I am looking in game stats engines hold with NN files by SF weaknesses and strengths if I am able to see that with statistics. A good example is the strong king attack from Rodent.

2.
I accept the opinion by programmers that the work on a NN file need a lot of time and is most of time boring. Read that often, in mails also. I have an other opinion if I am looking on the hard work John Stanback do here. A conflict with myself for different reasons.

3.
I wish me NN files with an easy licenc model.

Sooner or later ...
Neural Network is very new ...
Lifeblood programmers will create his own.
In the meantime we should not convict it.

Best
Frank

[Frank Quisinsky]

Hi there,

I changed the ratinglist:

JPG: https://www.amateurschach.de/t-ki/common/ki-ratings.jpg
TXT: https://www.amateurschach.de/t-ki/ratin ... atings.txt

Hope this is right?

To changed the databases (PGN & CBH) is a bit more work.
Will do that in the next days.

As next I will changed the engine link overview and some other parts of website.

To Günther, Conner, Peter and Martin ...
Thank for the discussion!

For Günther:
I start to create a database for the programmer of Bit-Genie (endgame problem).
This need a while.

Best
Frank

[chrisw]

Martin,

at first:
Nice to read you!

The deal?
I wrote that before ... I am just mate ... 50% to 50%, helpless is more right.

The reasons for myself:

1.
If I am looking in game stats engines hold with NN files by SF weaknesses and strengths if I am able to see that with statistics. A good example is the strong king attack from Rodent.

2.
I accept the opinion by programmers that the work on a NN file need a lot of time and is most of time boring.

No idea where you get that idea from. Training and testing NNs with different ideas, also integrating with search, is completely fascinating.

Read that often, in mails also. I have an other opinion if I am looking on the hard work John Stanback do here. A conflict with myself for different reasons.

3.
I wish me NN files with an easy licenc model.

Sooner or later ...
Neural Network is very new ...
Lifeblood programmers will create his own.
In the meantime we should not convict it.

Best
Frank

[connor_mcmonigle]

...

That looks like a reasonable solution to me.

[Frank Quisinsky]

Hello Chris,

Can't give names from programmers have that opinion and sent it by mail!

In the past different dislike to added Nalimov support, stupid and boring work after all I read.

But if you have fun with it ... good!
Do it and give us at last your Chess System Tal!



Tomorrow I have to drive to the beverage dealer.
You know nothing I like more, others like that.

Best
Frank

[Frank Quisinsky]

Hi Günther,

Bit-Genie 9.19 dev "endgame-problem":
I added a *.zip file in thread about bug-reports.

Best
Frank

[Frank Quisinsky]

Hi Conner,

you are speaking about CCRL.

Note:
CCRL and FGRL are my favorite rating systems.

What I do here with my tourney's isn't very important.
A bit fun in testing Wasp, looking games on my own systems in live mode.
The work with Klaus Wlotzka (Excel stats) is interesting for me.
John Stanback have fun with it.

Furthermore, I need strong material for my next FEOBOS project.
A good mix between correspondence, GM and comp-comp games.
I have different ideas to make FEOBOS better but this is a long time project (need around 2-3 years).
So I collected positions 4 moves after ECO codes formed.

Not a competition for CEGT, CCRL, FGRL, Ipman ... what others do.
Just fun in free time.

Fact is:
I like my tourney system with so many different and strong engines.
Information about it should be right.
The reason I am happy if I can get helps in questions I am not sure.

Keep up your good work around Seer and thanks for your engine.
Seer is on the way in TOP-10, very hard opponent for Wasp.



Many things with "NNSf" information are changed on my site.
Not all, working on it.

Best
Frank

[JohnWoe]

When it comes to Mayhem. Only Polyglot stuff is copy-pasted from Stockfish. Because.
Drum roll start ...
No point reinventing the wheel!
Drum roll end ...

When you implement a beautiful unique NNUE. Then support for all the boring stuff: threads, ponder, syzygy EGTB... Then I have reinvented Stockfish. That's boring.

[connor_mcmonigle]

When it comes to Mayhem. Only Polyglot stuff is copy-pasted from Stockfish. Because.
Drum roll start ...
No point reinventing the wheel!
Drum roll end ...

When you implement a beautiful unique NNUE. Then support for all the boring stuff: threads, ponder, syzygy EGTB... Then I have reinvented Stockfish. That's boring.

You link against the probing code from CFish and use weights from Stockfish effectively copy pasting the entire evaluation function as has been explained previously. If you can't see any value in experimenting with new techniques, then perhaps best is to just not write a chess engine. Stockfish will almost certainly always be stronger. Why are you reinventing the wheel?

NIH is only valid for highly specified problems with clearly optimal solutions. Several authors use external move generation libraries, EGTB probing code, polyglot probing code, etc. These are all problems to which NIH is applicable as a principle. The question of how to best map a chess position to a scalar is clearly not such a problem.