Proteus Chess SF - developing my Stockfish 14.1 derivative.

121/240... the level is so high.
I wouldn't be surprised if something like 122/240 can win the tournament

dorsz wrote:

121/240... the level is so high.
I wouldn't be surprised if something like 122/240 can win the tournament

Yes, but as you know, I prefer unbalanced pgn test suites likes DOTS5 ...much more thrillings

Restarted ProteusSFAurora.exp from scratch to make It unique according to Aurora playing style
as suggested by @sarona and to verify performance differences with ProteusSF.exp 008b Milkshake Edition

But I want to mod it much more before releasing, and it will take months, while the final playing style could be different from first tests and I will repeat all games of updated engines in the meawhile, not only my dev builds. So it's a work-in-progress round robin, too.

Wow! Already 2 nice wins by ProteusSF Aurora 009 dev (one with black) against Crystal 4.0 130622 by Joseph Ellis!

Hi Alex! could you post your settings in the engine uci

Or does Proteus SF play with the default settings?

Thanks in advance

Just released ProteusSF.exp 180622 update 598.5 MB with a lot of Playchess.com games by @igor and trained 8moves_v3.pgn DOWNLOAD IT HERE

Last edited by AlexChess on Sat Jun 18, 2022 5:23 pm; edited 2 times in total

Del

Fire_Fox wrote:

Hi Alex! could you post your settings in  the engine uci

Or does Proteus SF play with the default settings?

Thanks in advance

Overall options prevail to individual settings 8 CPUs - Tablebases Syzygy 6 man 510 edgames

Almost interesting as Candidates qualifications!  

@dorsz 's Swordfish 15.2 wins a game against ProteusSF Aurora 009 this one leads 3-1 against Andrea Manzo'a Shashchess 22 (until now)  

WATCH HERE

Completed also all 240 ProteusSF Aurora 009 dev games! Nice debut with 123/240 points! Winning 1 more game against Blue Marlin 15.2a, Corchess 3 NNUE 070622 and BrainLearn 15.2, 3 games against Shashchess 22 NNUE 130622 and 2 games against Crystal 4.0 130622, while losing 2 games only with Swordfish 15.2 and Shashchess 22.

Next to complete all 240 games: Kayra 1.4 to verify if it can do better.

WATCH HERE

In another forum I've read, Fisherov 0.98 have beaten ProteusSF RBE 008 with 57:43 (100 games 3' 3").

Doesn't Proteus like the playing style of Fisherov or do you try to make changes in actual/further versions to be more equal?

Fisherov is private so it is hard to answer your question.

Falireza wrote:

In another forum I've read, Fisherov 0.98 have beaten ProteusSF RBE 008 with 57:43 (100 games 3' 3").

Doesn't Proteus like the playing style of Fisherov or do you try to make changes in actual/further versions to be more equal?

If I could test Fisherov 0.98k... Since last pubblically avalable version is the obsolete 0.98b and no one want to pay to become betatester of a not GPLv3 compliant "private" Stockfish clone, nobody cares about that RBE 008 anyway is based on SugaR 2.50 ...I'd like to know results of 98K against 008b or I can do a fair match reversing side for each position for myself with 009 instead, if you have it

It is too bad a respected independent tester could not run matches with Fischerov against other engines - especially Eman.  It would give these results more credibility and would greatly increase the reputation of Fischerov. However, I don't see this happening any time soon.

Zeus and AI are not public either, but they are independently tested (oliprog). Maybe I do not agree with his test parameters (testing without increment), but I trust him to be non-biased.  Also when I read that an old ICCF veteran says that Zeus is his "go to" engine for analysis, I begin to respect the engine's capabilities.

There is none of this with Fischerov. Independent testing (in which the engine would not be shared publicly) would rectify this and give more evidence as to the engine's strength.

sarona wrote:

It is too bad a respected independent tester could not run matches with Fischerov against other engines - especially Eman.  It would give these results more credibility and would greatly increase the reputation of Fischerov. However, I don't see this happening any time soon.

Zeus and AI are not public either, but they are independently tested (oliprog). Maybe I do not agree with his test parameters (testing without increment), but I trust him to be non-biased.  Also when I read that an old ICCF veteran says that Zeus is his "go to" engine for analysis, I begin to respect the engine's capabilities.

There is none of this with Fischerov. Independent testing (in which the engine would not be shared publicly) would rectify this and give more evidence as to the engine's strength.

All Stockfish derivatives authors should honesty release their engines according to GPLv3 rules. I will no more test engines of hidden authors that do not respect Stockfish developers work. There are a lot of very skilled modders like @dorsz , @mehmet1921 , Ivan Ivec, Joseph Ellis and Andrea Manzo to friendly challenge, sharing ideas about our hobby and there is no reason to look for not GPLv3 compliant engines even not stronger than our mods and that add nothing to chess enthusiasts world (especially private ones) ! Instead, I will restart testing FREE original engines availables for all operating systems thanks to their open source nature: Berserk 9, Koivisto 8.13 and Seer 2.5.0 that have almost reached Stockfish strength!!! Great respect for them!

It's been a long time I've never gave all of the Fisherov test results my thumb's up or even take a look at results anymore..

It's not because I don't appreciate the tester or the man whos posted it, it was just simply THE WAY they developes their engine against another engines and then matched them in a battle..

For me.. when we're going to race in a .. Lets say.. a drag race of ¼ mile between 2 race cars, then all participants should have a chance to look up what is under the hood of their rival's cars..

And not the other way arround, where they can see ( and perhaps study it ) what's in our hood, while we can't see theirs..

That's what happened when an engine which hidding their GPL licence compete against an open source engine..

Simply not fair enough !!

You know how I feel about private derivatives when we talked about Aurora. My opinion has not changed.  Nobody is obliged to release their derivative because it exists.  Kudos to those that do and respect the GPLv3.

Proteus is using somewhat modified source from SugaR 2.50 (although unmodified SugaR AI 2.50 won a CEDR tournament earlier this month by a point and a half LOL!   ).

So v008a is a bit dated. That should not happen with v009 onwards.  The Fischerov-Proteus tester was probably not using experience. There were no books (good idea) and he used increment (even better).  So Proteus should be the underdog anyway. Seven of your losses revolved around ECO C11 (French Defense - classical line).

In another tournament EMAN 8.03 finished fifth in a very short tournament with Fischerov 0.98k winning. So judge for yourself these results.

sarona wrote:

You know how I feel about private derivatives when we talked about Aurora. My opinion has not changed.  Nobody is obliged to release their derivative because it exists.  Kudos to those that do and respect the GPLv3.

Proteus is using somewhat modified source from SugaR 2.50 (although unmodified SugaR AI 2.50 won a CEDR tournament earlier this month by a point and a half LOL!   ).

Code:

https://chessengines.blogspot.com/2022/06/sugar-ai-250-wins-1-league-cedr-edition.html

So v008a is a bit dated. That should not happen with v009 onwards.  The Fischerov-Proteus tester was probably not using experience. There were no books (good idea) and he used increment (even better).  So Proteus should be the underdog anyway. Seven of your losses revolved around ECO C11 (French Defense - classical line).

In another tournament EMAN 8.03 finished fifth in a very short tournament with Fischerov 0.98k winning. So judge for yourself these results.

Completely agree with you if the engine is really private, not released only to PAYING testers ...or shared by error .

He also mix random engines with a completely different strength: Stockfish 15 against Pedone 3.1 and Raubfisch X48e, so these tests are completely meaningless and useless for me. Like testing a Ferrari against a Fiat 500.

For that reason I'm trying to rationalize my method of testing, for example repeating each position with reverse colors for a fair result.

I tested. chukaro 4 bin book and proteus sf RBE 008b engine . good results actually 3800 + retting opening book .. and finally winning for advantage 99%

Krunal wrote:

I tested.  chukaro 4 bin book and proteus sf RBE 008b engine .  good results actually 3800 + retting opening book .. and finally winning for advantage 99%

WOW!!!! And it's an important result for me since it is an independent test.

Shashchess 22 has just won a game against Kayra 1.4, probably compromising @Mehmet1921 leadership over both ProteusSF!

My engines usually fears Kayra, but not Shashchess, is the opposite between them.

This is a demonstration that all Stockfish derivatives (even with light fine tuning like first ProteusSF Aurora 009 build that I'm studying to understand weaknesses and forces) play very differently from Stockfish dev

Also Corchess 3 070622, BrainLearn 17 and Stockfish 070622 have won a game against Kayra 1.4!

WATCH HERE

All games in PGN and crosstable updates only when Kayra 1.4 has completed his 240 games. 40° Celsius here

RounDrobin Stockfish-dev against all GPLv3 derivatives just updated with all Kayra 1.4 results (648 games).

WATCH HERE

Playing all 240 Swordfish 15.2 games right now.

Hello Alex chess my just questions..new Proteus SF exp file 180622 download successfully but ..how do I replace it? For example engine Proteus SF RBE 008 b > manage chess engine > set options > finally experience file new how to add.? Already old file?

A new file has been downloaded. But how do I add it? Droidfish app for android..already old experience file .?

Krunal wrote:

Hello Alex chess my just questions..new      Proteus SF exp file  180622 download successfully but ..how do I replace it?                     For example engine Proteus SF RBE 008 b >  manage chess engine > set options > finally experience file new how to add.?  Already old file?

You must overwrite the existing one, copying the new over it on the same folder.

Last edited by AlexChess on Tue Jun 21, 2022 3:58 pm; edited 1 time in total

FIRE ON BOARD!  

Stockfish dev 070622 tries to take an advantage over @dorsz 's Swordfish 15.2, but the last one reacts on next game and re-takes his point on same position with reverse colors!

WATCH HERE

One more demonstration that reversing sides on each position using PGN test suites is the right way to fairly test engines.

Customized bin books to be used only on challenges, not on developing tests like this one

THE 2 WON GAMES IN PGN FORMAT