NNUE Research Project

[Rebel]

Ok good to know this was a good sv net. Thanks.

To clarify, it will be helpful to call those Minic2.46 to 2.53 MinicNNUE+named net, if a net is being used.

I only really "assume" nnue technology starting from Minic 3.00 where the implementation was switchd to Seer one and architecture is not the same as SF at all. Now in the next Minic 3.05 to come, supported net has even diverged from Seer architecture.

To go back on the very high similarity here, I think we already discuss that back in September. We conclude evaluation, so here net in used, is much more important than search in similarity testing. So using here a SF data based net is clearly leading to high similarity. On the contrary, looking at a Minic data net (but still using the same architecture as the SF one) namely nascent nutrient, we get a smaller similarity.

So whatever the search is, seeing nets that have high similarity with SF probably indicates SF data were used to build the net.

I updated the page.

[David Carteau]

I updated the page.

It would be great if you could also update the Orion 0.7 part (see my previous comment in this thread : in short, rename "Orion 0.7" with "Orion 0.7 NNUE" and I confirm that the network used by this experimental and not official version of Orion was the SF12's one, hence the high expected similarity )

A great thank !

[RubiChess]

NNUE Research Project
March 10, 2021

It´s generally known by now similarity testing on moves does not work with NNUE nets. On this page we will try to research if it is not possible using other methods. One method is to calculate the Root-mean-square deviation (or RMS) of the scores instead of moves as after all NNUE is a set of scores. We will present data and the source code for discussion.

Let´s start at the beginning of NNUE in the summer of 2020 the starting point of the NNUE revolution when the Stockfish team implemented the Sergio nets. Our first goal is to measure the stability of the RMS of Stockfish NNUE nets. From the Sergio nets we calculate the RMS of the very first 3 nets (july) and the last 3 (september) and compare the RMS with the final SF12 net, see table one. In table two the nets between SF12 and SF13 are compared plus 5 nets after the release of SF13.

....

http://rebel13.nl/home/nnue.html

Observation/remark #2 looks strange. Talks about Igel-2.7/2.8 but then switches to Fat Fritz 2 instead of Igel-2.9 (which looks different from <=2.8).

Regards, Andreas

[Guenther]

I updated the page.

It would be great if you could also update the Orion 0.7 part (see my previous comment in this thread : in short, rename "Orion 0.7" with "Orion 0.7 NNUE" and I confirm that the network used by this experimental and not official version of Orion was the SF12's one, hence the high expected similarity :))

A great thank !

Yes. And you were very clear about this and also linked to the relevant infos from your website.
But it seems not all have read it...
http://talkchess.com/forum3/viewtopic.php?f=2&t=74828

[dkappe]

NNUE Research Project
March 10, 2021

It´s generally known by now similarity testing on moves does not work with NNUE nets. On this page we will try to research if it is not possible using other methods. One method is to calculate the Root-mean-square deviation (or RMS) of the scores instead of moves as after all NNUE is a set of scores. We will present data and the source code for discussion.

Let´s start at the beginning of NNUE in the summer of 2020 the starting point of the NNUE revolution when the Stockfish team implemented the Sergio nets. Our first goal is to measure the stability of the RMS of Stockfish NNUE nets. From the Sergio nets we calculate the RMS of the very first 3 nets (july) and the last 3 (september) and compare the RMS with the final SF12 net, see table one. In table two the nets between SF12 and SF13 are compared plus 5 nets after the release of SF13.

....

http://rebel13.nl/home/nnue.html

Observation/remark #2 looks strange. Talks about Igel-2.7/2.8 but then switches to Fat Fritz 2 instead of Igel-2.9 (which looks different from <=2.8).

Regards, Andreas

Igel 2.7 and 2.8 we’re running night nurse. With 2.9, it was running a net trained on igel’s HCE. This is covered in posts here and on voffka’s github.

[Ferdy]

Regarding the questions, I compiled a download with all the epd files at http://rebel13.nl/nnue-epd.7z

Contents:

Code: Select all

11-03-2021  19:58    <DIR>          .
11-03-2021  19:58    <DIR>          ..
11-03-2021  19:54    <DIR>          ff2
04-03-2021  08:59           647.686 Nemorino.epd
09-03-2021  22:18           644.556 Orion_0.8.epd
08-03-2021  12:18           649.858 SF12-Igel-270.epd
08-03-2021  12:35           649.635 SF12-Igel-280.epd
08-03-2021  21:18           652.073 SF12-Igel-290.epd
08-03-2021  12:53           652.728 SF12-Minic.epd
08-03-2021  13:11           653.371 SF12-napping-nexus.epd
08-03-2021  13:29           650.594 SF12-nascent-nutrient.epd
09-03-2021  11:03           652.710 SF12-Orion_0.7.epd
08-03-2021  13:46           652.583 SF12-sf-0c6fc5ef48e1.epd
08-03-2021  17:44           652.711 SF12-sf-516f5b95189a.epd
08-03-2021  17:27           652.513 SF12-sf-dd0c4c630f7e.epd
09-03-2021  09:40           652.457 SF12-sv-20200720-1017.epd
09-03-2021  10:00           652.411 SF12-sv-20200721-0909.epd
09-03-2021  10:20           652.651 SF12-sv-20200721-1432.epd
09-03-2021  10:38           652.882 SF12-sv-20200906-1046.epd
09-03-2021  10:56           652.892 SF12-sv-20200908-1733.epd
09-03-2021  11:15           652.759 SF12-sv-20200914-1520.epd
08-03-2021  11:43           652.698 SF12.epd
08-03-2021  21:35           651.330 SF13-Igel-290.epd
08-03-2021  23:30           652.440 SF13-Rubi-2.01.epd
09-03-2021  09:06           651.983 SF13-sf-6b7a4192c303.epd
09-03-2021  09:26           651.878 SF13-sf-94816594b327.epd
08-03-2021  12:01           652.459 SF13.epd

Remarks:
1. nets are tested as much as possible with SF12 and SF13.
2. Nemorino and Orion 0.8 are the exceptions since they have a different file format.
3. The Orion 0.7 version has the exact SF12 net.
4. the folder ff2 contains the epd's of sf12, 13 and ff2 at 100ms, 250 and 500ms, made by someone else on a different pc.
5. epd's labelled with "sv" are the tested "sergio" nets.
6. epd's labelled with "sf" are Stockfish nets.

Under folder ff2:
FF2-100ms.epd is the output when ff2 engine uses the ff2 net?
What is SF12-100ms.epd?

[RubiChess]

Observation/remark #2 looks strange. Talks about Igel-2.7/2.8 but then switches to Fat Fritz 2 instead of Igel-2.9 (which looks different from <=2.8).

Regards, Andreas

Igel 2.7 and 2.8 we’re running night nurse. With 2.9, it was running a net trained on igel’s HCE. This is covered in posts here and on voffka’s github.

I'm not interested in any fetish giving female (or animal) names to some network files, just mentioned that the text drifts away from Igel to the (not related?) Fat Fritz 2. Maybe Rebel just did some typo or cut-and-paste error?

[dkappe]

Igel 2.7 and 2.8 we’re running night nurse. With 2.9, it was running a net trained on igel’s HCE. This is covered in posts here and on voffka’s github.

I'm not interested in any fetish giving female (or animal) names to some network files, just mentioned that the text drifts away from Igel to the (not related?) Fat Fritz 2. Maybe Rebel just did some typo or cut-and-paste error?

What is a “fetish giving female name?” Is this something you want to share?

[Guenther]

Igel 2.7 and 2.8 we’re running night nurse. With 2.9, it was running a net trained on igel’s HCE. This is covered in posts here and on voffka’s github.

I'm not interested in any fetish giving female (or animal) names to some network files, just mentioned that the text drifts away from Igel to the (not related?) Fat Fritz 2. Maybe Rebel just did some typo or cut-and-paste error?

What is a “fetish giving female name?” Is this something you want to share?

You have serious comprehension problems. Did you try at least to completely read Andreas first post?
(We know your main expertise is a kind of childish trolling, but still...)

Here is the part you cut off and 'replied' too. Ofc you have to look at remark #2 otherwise it makes no sense to reply at all...

RubiChess wrote:
Fri Mar 12, 2021 12:35 pm
Observation/remark #2 looks strange. Talks about Igel-2.7/2.8 but then switches to Fat Fritz 2 instead of Igel-2.9 (which looks different from <=2.8).

Regards, Andreas

[RubiChess]

What is a “fetish giving female name?” Is this something you want to share?

I don't want to share anything with you. Learned my lesson about you and your arrogant attitude long time ago.
"The joy of seeing Bagatur (or whatever it’s called) play asymptote will be boundless".