FCP Tourney-KI started January 02nd, 2022

[Frank Quisinsky]

Hi there,

but what I can do is ...
To create an event from my mistake ...

I can test after "run-6" with a g-run ... Halogen 10.23.11 NN dev with endgame databases.
So I have both in my list and we can see the differents (with and without syzygy).

If any person like to see that I will do it.

Best
Frank

[Frank Quisinsky]

Code: Select all

FCP Tourney-KI
run-6 after round 10 out of 30


         Name                       Games     Won     Draw     Lose       Pts         S-B         %

01. Stockfish 020622 NN dev       :   400  :  233+  :  167=  :    0-  :   316.5  :  60522.75  :  79.13%
02. Dragon 3 NN (Komodo)          :   400  :  226+  :  174=  :    0-  :   313.0  :  60077.25  :  78.25%
03. Koivisto 8.9 NN               :   400  :  171+  :  220=  :    9-  :   281.0  :  52856.00  :  70.25%
04. Fire 8 MC.3 NN                :   400  :  165+  :  229=  :    6-  :   279.5  :  52984.25  :  69.88%
05. SlowChess Blitz 2.9 NN        :   400  :  167+  :  222=  :   11-  :   278.0  :  52113.50  :  69.50%
06. Berserk 9 NN dev3             :   400  :  164+  :  222=  :   14-  :   275.0  :  51538.00  :  68.75%
07. Revenge 2.0 NN                :   400  :  151+  :  233=  :   16-  :   267.5  :  49977.75  :  66.88%
08. rofChade 3.0 NN               :   400  :  157+  :  218=  :   25-  :   266.0  :  49824.25  :  66.50%
09. RubiChess 20220223 NN         :   400  :  137+  :  243=  :   20-  :   258.5  :  48118.50  :  64.63%
10. Seer 2.5.0 NN                 :   400  :  133+  :  233=  :   34-  :   249.5  :  46294.50  :  62.38%
11. Arasan 23.3 NN                :   400  :  108+  :  260=  :   32-  :   238.0  :  44374.25  :  59.50%
12. Minic 3.22 NN                 :   400  :  101+  :  257=  :   42-  :   229.5  :  42467.00  :  57.38%
13. Igel 3.1.0 NN                 :   400  :   86+  :  276=  :   38-  :   224.0  :  41890.00  :  56.00%
14. Halogen 10.23 NN dev          :   400  :  103+  :  225=  :   72-  :   215.5  :  39103.25  :  53.88%
15. Ethereal 13.07                :   400  :   86+  :  257=  :   57-  :   214.5  :  39824.25  :  53.63%
16. Lc0 0.28.2 752187 CPU         :   400  :   86+  :  249=  :   65-  :   210.5  :  39249.50  :  52.63%
17. Clover 3.1 NN                 :   400  :   82+  :  253=  :   65-  :   208.5  :  38331.50  :  52.13%
18. Wasp 5.54 NN dev              :   400  :   81+  :  254=  :   65-  :   208.0  :  38667.75  :  52.00%
19. Nemorino 6.09 NN dev          :   400  :   84+  :  248=  :   68-  :   208.0  :  38375.50  :  52.00%
20. Booot 7.0 NN dev              :   400  :   69+  :  265=  :   66-  :   201.5  :  37365.25  :  50.38%
21. Uralochka 3.36c NN            :   400  :   72+  :  247=  :   81-  :   195.5  :  36094.00  :  48.88%
22. Tucano 10.00 NN               :   400  :   59+  :  270=  :   71-  :   194.0  :  35663.75  :  48.50%
23. Rebel 15x2 NN                 :   400  :   72+  :  243=  :   85-  :   193.5  :  35486.00  :  48.38%
24. Fritz 18 NN (Ginkgo)          :   400  :   67+  :  231=  :  102-  :   182.5  :  33032.00  :  45.63%
25. Zahak 10.0 NN                 :   400  :   57+  :  225=  :  118-  :   169.5  :  30517.25  :  42.38%
26. Velvet 3.3.0 NN               :   400  :   47+  :  239=  :  114-  :   166.5  :  30275.50  :  41.63%
27. Xiphos 0.6                    :   400  :   57+  :  214=  :  129-  :   164.0  :  29760.75  :  41.00%
28. Combusken 2.0.0               :   400  :   48+  :  222=  :  130-  :   159.0  :  28955.25  :  39.75%
29. Weiss 2.0                     :   400  :   50+  :  217=  :  133-  :   158.5  :  29178.25  :  39.63%
30. Hiarcs 15.1                   :   400  :   37+  :  223=  :  140-  :   148.5  :  27419.25  :  37.13%
31. Defenchess 2.3 dev            :   400  :   34+  :  229=  :  137-  :   148.5  :  27340.00  :  37.13%
32. Schooner 2.2 XB               :   400  :   32+  :  233=  :  135-  :   148.5  :  26991.00  :  37.13%
33. Laser 1.7                     :   400  :   36+  :  219=  :  145-  :   145.5  :  26495.50  :  36.38%
34. Bit-Genie 9.19 dev            :   400  :   45+  :  196=  :  159-  :   143.0  :  25975.75  :  35.75%
35. Fizbo 2.0                     :   400  :   46+  :  194=  :  160-  :   143.0  :  25896.75  :  35.75%
36. DanaSah 9.0 NN                :   400  :   25+  :  233=  :  142-  :   141.5  :  26276.50  :  35.38%
37. Shredder 13                   :   400  :   36+  :  208=  :  156-  :   140.0  :  25417.25  :  35.00%
38. Stash 33.0                    :   400  :   33+  :  205=  :  162-  :   135.5  :  24779.50  :  33.88%
39. Marvin 5.2.0 NN               :   400  :   32+  :  205=  :  163-  :   134.5  :  24414.25  :  33.63%
40. Chiron 5                      :   400  :   32+  :  198=  :  170-  :   131.0  :  23225.75  :  32.75%
41. Expositor 2WN29 NN            :   400  :   19+  :  192=  :  189-  :   115.0  :  21440.75  :  28.75%

After 33,33% of tournament!
Round 10 out of 30!

Current ratings, v6.10
https://www.amateurschach.de/t-ki/common/ki-ratings.jpg



Best
Frank

[connor_mcmonigle]

Code: Select all

FCP Tourney-KI
run-6 after round 10 out of 30


         Name                       Games     Won     Draw     Lose       Pts         S-B         %

01. Stockfish 020622 NN dev       :   400  :  233+  :  167=  :    0-  :   316.5  :  60522.75  :  79.13%
02. Dragon 3 NN (Komodo)          :   400  :  226+  :  174=  :    0-  :   313.0  :  60077.25  :  78.25%
03. Koivisto 8.9 NN               :   400  :  171+  :  220=  :    9-  :   281.0  :  52856.00  :  70.25%
04. Fire 8 MC.3 NN                :   400  :  165+  :  229=  :    6-  :   279.5  :  52984.25  :  69.88%
05. SlowChess Blitz 2.9 NN        :   400  :  167+  :  222=  :   11-  :   278.0  :  52113.50  :  69.50%
06. Berserk 9 NN dev3             :   400  :  164+  :  222=  :   14-  :   275.0  :  51538.00  :  68.75%
07. Revenge 2.0 NN                :   400  :  151+  :  233=  :   16-  :   267.5  :  49977.75  :  66.88%
08. rofChade 3.0 NN               :   400  :  157+  :  218=  :   25-  :   266.0  :  49824.25  :  66.50%
09. RubiChess 20220223 NN         :   400  :  137+  :  243=  :   20-  :   258.5  :  48118.50  :  64.63%
10. Seer 2.5.0 NN                 :   400  :  133+  :  233=  :   34-  :   249.5  :  46294.50  :  62.38%
11. Arasan 23.3 NN                :   400  :  108+  :  260=  :   32-  :   238.0  :  44374.25  :  59.50%
12. Minic 3.22 NN                 :   400  :  101+  :  257=  :   42-  :   229.5  :  42467.00  :  57.38%
13. Igel 3.1.0 NN                 :   400  :   86+  :  276=  :   38-  :   224.0  :  41890.00  :  56.00%
14. Halogen 10.23 NN dev          :   400  :  103+  :  225=  :   72-  :   215.5  :  39103.25  :  53.88%
15. Ethereal 13.07                :   400  :   86+  :  257=  :   57-  :   214.5  :  39824.25  :  53.63%
16. Lc0 0.28.2 752187 CPU         :   400  :   86+  :  249=  :   65-  :   210.5  :  39249.50  :  52.63%
17. Clover 3.1 NN                 :   400  :   82+  :  253=  :   65-  :   208.5  :  38331.50  :  52.13%
18. Wasp 5.54 NN dev              :   400  :   81+  :  254=  :   65-  :   208.0  :  38667.75  :  52.00%
19. Nemorino 6.09 NN dev          :   400  :   84+  :  248=  :   68-  :   208.0  :  38375.50  :  52.00%
20. Booot 7.0 NN dev              :   400  :   69+  :  265=  :   66-  :   201.5  :  37365.25  :  50.38%
21. Uralochka 3.36c NN            :   400  :   72+  :  247=  :   81-  :   195.5  :  36094.00  :  48.88%
22. Tucano 10.00 NN               :   400  :   59+  :  270=  :   71-  :   194.0  :  35663.75  :  48.50%
23. Rebel 15x2 NN                 :   400  :   72+  :  243=  :   85-  :   193.5  :  35486.00  :  48.38%
24. Fritz 18 NN (Ginkgo)          :   400  :   67+  :  231=  :  102-  :   182.5  :  33032.00  :  45.63%
25. Zahak 10.0 NN                 :   400  :   57+  :  225=  :  118-  :   169.5  :  30517.25  :  42.38%
26. Velvet 3.3.0 NN               :   400  :   47+  :  239=  :  114-  :   166.5  :  30275.50  :  41.63%
27. Xiphos 0.6                    :   400  :   57+  :  214=  :  129-  :   164.0  :  29760.75  :  41.00%
28. Combusken 2.0.0               :   400  :   48+  :  222=  :  130-  :   159.0  :  28955.25  :  39.75%
29. Weiss 2.0                     :   400  :   50+  :  217=  :  133-  :   158.5  :  29178.25  :  39.63%
30. Hiarcs 15.1                   :   400  :   37+  :  223=  :  140-  :   148.5  :  27419.25  :  37.13%
31. Defenchess 2.3 dev            :   400  :   34+  :  229=  :  137-  :   148.5  :  27340.00  :  37.13%
32. Schooner 2.2 XB               :   400  :   32+  :  233=  :  135-  :   148.5  :  26991.00  :  37.13%
33. Laser 1.7                     :   400  :   36+  :  219=  :  145-  :   145.5  :  26495.50  :  36.38%
34. Bit-Genie 9.19 dev            :   400  :   45+  :  196=  :  159-  :   143.0  :  25975.75  :  35.75%
35. Fizbo 2.0                     :   400  :   46+  :  194=  :  160-  :   143.0  :  25896.75  :  35.75%
36. DanaSah 9.0 NN                :   400  :   25+  :  233=  :  142-  :   141.5  :  26276.50  :  35.38%
37. Shredder 13                   :   400  :   36+  :  208=  :  156-  :   140.0  :  25417.25  :  35.00%
38. Stash 33.0                    :   400  :   33+  :  205=  :  162-  :   135.5  :  24779.50  :  33.88%
39. Marvin 5.2.0 NN               :   400  :   32+  :  205=  :  163-  :   134.5  :  24414.25  :  33.63%
40. Chiron 5                      :   400  :   32+  :  198=  :  170-  :   131.0  :  23225.75  :  32.75%
41. Expositor 2WN29 NN            :   400  :   19+  :  192=  :  189-  :   115.0  :  21440.75  :  28.75%

After 33,33% of tournament!
Round 10 out of 30!

Current ratings, v6.10
https://www.amateurschach.de/t-ki/common/ki-ratings.jpg



Best
Frank

Just a minor correction, but for consistency, "Fire 8 MC" should be labeled as "NNSf" by your convention.

[Frank Quisinsky]

Hi Conner,

same for DanaSah 9.0 NN, same for Orion 0.8 NN (also after the explanation from the programmer of Orion).

Not clear for me!!

I am not the referee of programmers work.
Not with the knowledge I have about it.
And If I have the knowledge about it ... never I would gave such an information if the programmer have an other opinion!

Sorry, in any case it's not clear for me I can't give this information.

On my detail site I wrote:
https://www.amateurschach.de/main/_fcp-tourney-ki.htm

Code: Select all

Rules, Note 2:
Please be informed that several programmers used NN (Neural Network) files based on Stockfish!
For more information have a look on the sites by the programmers!
Since March 09th, 2022 you can detect these group of engines with my hint "NNSf".

NNSf = Neural Network Stockfish
NNRe = Neural Network Rebel

Controversial is Fire 8 MC.3 NN.
In my humble opinion the programmer developed an own Neural Network.

Some other available engines (for the moment not tested by me) used NNSf.
Keeping an overview is difficult.

To play police-men ... this part can other do.
Speaking from childrens-club of different chess programmers!

Start on github pages with all the ideas inside in available engines without to give any information about it.
Again, not in my interest!

Best
Frank

[Frank Quisinsky]

Hi there,

but possible is ...
Each engine have many opponents ... min. 40!

To download my work here:
https://www.amateurschach.de/main/_download-ki.htm

And to public the work on an other website with engines like or unlike only.
Very easy to do that with Bayesian, EloStat or Ordo.
All what I do is available here ...

https://www.amateurschach.de/main/_fcp-tourney-ki.htm

I have no problems with it.
Not with my own data based on the work the programmers do.

Note 1:
After run-8, ratings of newer engines are private available.
So, I must not answere again and again of such messages you wrote.

Note 2:
I wonder that others, working on ratings, not get so often such hints.

Example:
Fire is in all rating systems!
And not only Fire ... I saw Houdini, FatFritz, private available engines and also this or that NNSf engine.
Example: Nemorino 6.0

To discuss about the work different people to with ratings is not right!
We gave a lot of money for hardware, electricity and much more important ... a lot of time ... to do this
for users and programmers.

Best
Frank

[Frank Quisinsky]

Engines analyses the same 41.614 openings (FEOBOS).
Similarties ...

Fire do a great job, all the time here.
I can't see that anything in Fire is a problem!

To hustle against Norman Schmidt (he do a lot things for the community) is highly questionable and not in my interest.
The result is group building between the programmers and after all my experience in the past is exactly this one not good for our nice hobby.

My opinion here is very clear and should be easy to understand.

Go Fire go ...
Same for all others in my list.

Best
Frank

[pedrox]

same for DanaSah 9.0 NN, same for Orion 0.8 NN (also after the explanation from the programmer of Orion).

First of all thanks for testing DanaSah, it is possible that without people testing it I would have passed on version 2. I am happy that DanaSah is at the top of the engines with a more aggressive style.

I read the talkchess forum often. It is easy for me to answer questions and doubts. I have already seen an earlier post where a programmer questions DanaSah's work. I really felt a great displeasure and had doubts whether to answer or not, I let it go, better not to answer in the heat of the moment.

As for DanaSah version 9 using a nnue network, I have more or less explained the process I have followed to create the network. This has not consisted of taking one of the networks created for Stockfish. I think it is pointless to question whether for example DanaSah should carry the NNSf label. DanaSah does not use any network created by another programmer or person, nor does it use any data.

I do not know who imposed the idea that each engine should use its own evaluations to create the network and could not use evaluations from other engines, this was imposed as a must. Who decided it and why? If someone wants to organize a tournament under those circumstances I am fine with it. But I don't feel the need for someone to tell me how to make my engine.

DanaSah does not use DanaSah evaluations to create the network, in fact it does not use any engine evaluations. To create the initial network I decided to use a database, at least I needed to get about 300 million positions, so I decided to use the CCRL 40/4 database which had 2.5 million games. In this case the network is created not on the basis of the evaluation of the position by an engine but on the basis of the final result of the game. The network obtained and that I made public is a network that is already stronger than DanaSah's original evaluation, with a totally positional style and in which the network almost does not distinguish the value of the pieces and with all that it beats my original evaluation.

In the second step, I need the network to learn better the value of the pieces, for that I need to generate at least another 300 million positions from that first network, to generate the positions I have decided to use the nodchip generator. The 300 million positions in depth 5 were created during about 7 days on an AMDFX8300. The nodchip generator is not used to get an advantage over other ways, for example to generate sets in cutechess, I have used it because in my case the process is 6 times faster, otherwise I would have waited 6 weeks on my computer. It is even possible to take the data files generated by nodchip and have them evaluated by your engine (withouth nodchip), but again it is 6 times more work in my case and I don't really see that I have done anything wrong and that it will change anything.

In the third step to generate the DanaSah network, I generated another 300 million positions at depth 9, it took 3 weeks of processing on my AMDFX8300.

Of course this whole process is not something that came out the first time. It is a trial and error process that took about 6 months. Largely because the latest versions of nodchip made changes that caused the networks to not learn if the correct parameters were not set.

DanaSah's network is about 175 points weaker than Stockfish 12 net, I also created a network that is 125 points better but uses stockfish data, so I decided not to use it. Not everything is Elo.

While my work is being questioned for using nodchip, Stockfish decided that they are going to use the evaluation data generated by Leela. If I had been the one to decide that my engine would not have been tested. If I am being questioned for using nodchip, why isn't there a bigger protest from other programmers with Stockfish? Of course I have no problem with Stockfish's decision.

Of course, I believe that in chess computer you have to cut and avoid the use of clones, people who simply compile an engine and change 4 random values without any purpose and then pass the work off as original. But from there, to question the work of other engines (low cost engines) I think it should not be allowed. More when the programmers who say so have taken everything from other engines. They are allowed to study every line of Stockfish, take the idea and apply it to their engines, and even use code from other programmers as when accessing the tablebases and another. Ah but if I use genfens from nodchip my engine is already low cost!

[Frank Quisinsky]

Hi Pedro,

at first, thanks for your message.

To your engine DanaSah:
I am happy with your work and the nice style your engine have.
From my view all is wonderful.

With more easy words ...
Danasah = A good to have engine!

You renember on Thomas Meyer?
I gave him my older Winboard site ...
After I remember Thomas wrote often that DanaSah will have a great feature.

Shortly, why I have wrote that:
In a discussion with the programmer of Seer (around the time I added the NN"Sf" infos in engine names) he wrote that DanaSah is unclear for him. I read that first time and search information about it without to find anything. Never I changed the status for DanaSah from NN to NNSf.

I can give the status NNSf only if I have absolutly clear information about it, from the programmer of the engine.

And here I do a mistake:
For some weeks I gave Fire the status NNSf after a discussion with the programmer of Seer.
I gave Orion the status NNSf (comes from myself, not sure with Orion all the time).

I changed back without "Sf" after I read what the programmers of Fire and Orion have written about his own work.

I am thinking ...
I have not the knowledge to give such a status as non programmer.
Furthermore, information comes by the programmers of these engines are most important.
Not what other programmers thinking about a work from an other programmer.

I know that different programmers like to play police-men and have problems with "to accept".
I know that different programmers build groups and try to provoke more strenghts in his opinions in such groups.

I got in the last months many copies what programmers wrote to other programmers about my person.
All this dirty and wirrwarr material I will not have in my mail program.
I delete it without to send an answer.

I have enough from it from Winboard times, Arena times with all the people have problems with Winboard, UCI, Arena (often commercial companies).

Have nothing to do with your nice program.
But maybe you understand my point.

I try to organice a nice chess tournament, very complicated because it is a continous tourney. Such a tourney-mode never I try out before, very complicated in realization.

Thanks again and have a nice evening!
Thanks for DanaSah ...

Go DanaSah go ...

Best
Frank

[connor_mcmonigle]

Hi Conner,

same for DanaSah 9.0 NN, same for Orion 0.8 NN (also after the explanation from the programmer of Orion).

Not clear for me!!

I am not the referee of programmers work.
Not with the knowledge I have about it.
And If I have the knowledge about it ... never I would gave such an information if the programmer have an other opinion!

Sorry, in any case it's not clear for me I can't give this information.

On my detail site I wrote:
https://www.amateurschach.de/main/_fcp-tourney-ki.htm

Code: Select all

Rules, Note 2:
Please be informed that several programmers used NN (Neural Network) files based on Stockfish!
For more information have a look on the sites by the programmers!
Since March 09th, 2022 you can detect these group of engines with my hint "NNSf".

NNSf = Neural Network Stockfish
NNRe = Neural Network Rebel

Controversial is Fire 8 MC.3 NN.
In my humble opinion the programmer developed an own Neural Network.

Some other available engines (for the moment not tested by me) used NNSf.
Keeping an overview is difficult.

To play police-men ... this part can other do.
Speaking from childrens-club of different chess programmers!

Start on github pages with all the ideas inside in available engines without to give any information about it.
Again, not in my interest!

Best
Frank

The author of Fire did not develop his own network. It's not the same as Danasah and Orion. This is not really a matter of opinion. Here is a link to the network used in Fire, directly from Stockfish: https://tests.stockfishchess.org/api/nn ... fd0df.nnue. Hopefully that clears up any confusion.

[Guenther]

Code: Select all

Controversial is Fire 8 MC.3 NN.
In my humble opinion the programmer developed an own Neural Network.

Some other available engines (for the moment not tested by me) used NNSf.
Keeping an overview is difficult.

This is very wrong, Fire 8.3 simply reuses a renamed early Sergio Vieri net trained for SF!
The reason for renaming is obvious...
Actually someone who needs to rename other work to appear 'innocent' in the eyes of uninformed
persons should be a persona non grata, especially with all other things done in the past too.

If you need a hard proof I can upload both nn files the renamed so called 'raptor'.bin (fake name)
and the original nn file from Sergio Vieri and a binary comparison, which shows them to be identical.

Edit:

This should be sufficient:
https://tests.stockfishchess.org/nns?page=10
https://tests.stockfishchess.org/api/nn ... fd0df.nnue

Code: Select all

 Datei: nn-82215d0fd0df.nnue
  CRC-32: de239f9f
   SHA-1: 80f931862d61aa7898bf02254c42b721d18ee3b4
 SHA-256: 82215d0fd0df3cde17fc47ac024c5b5736823ec390200abdd705c44660c69ee4
 SHA-512: 5ac98071de7fd67c2a384040f4d7f4b803ce9402eb5fd1f7ebdabef28ab44c39c74d66e16e80445f7d368bab9e155cd32e501a34e544927bb9a7d5c8f36e2abe

https://github.com/FireFather/fire-NN/b ... raptor.bin

Code: Select all

   Datei: raptor.bin
  CRC-32: de239f9f
   SHA-1: 80f931862d61aa7898bf02254c42b721d18ee3b4
 SHA-256: 82215d0fd0df3cde17fc47ac024c5b5736823ec390200abdd705c44660c69ee4
 SHA-512: 5ac98071de7fd67c2a384040f4d7f4b803ce9402eb5fd1f7ebdabef28ab44c39c74d66e16e80445f7d368bab9e155cd32e501a34e544927bb9a7d5c8f36e2abe

also used here:
https://github.com/FireFather/sf-nnue/r ... 3-2022-aio
https://github.com/FireFather/sf-nnue/b ... val/nn.bin

Code: Select all

   Datei: nn.bin
  CRC-32: de239f9f
   SHA-1: 80f931862d61aa7898bf02254c42b721d18ee3b4
 SHA-256: 82215d0fd0df3cde17fc47ac024c5b5736823ec390200abdd705c44660c69ee4
 SHA-512: 5ac98071de7fd67c2a384040f4d7f4b803ce9402eb5fd1f7ebdabef28ab44c39c74d66e16e80445f7d368bab9e155cd32e501a34e544927bb9a7d5c8f36e2abe