Proteus Chess SF - Developing my Stockfish derivative

All about chess engines

Moderators: AlexChess, TedSummers

Locked
User avatar
AlexChess
Posts: 93
Joined: Thu Nov 04, 2021 5:55 am
Location: Rome, Italy
Contact:

Proteus Chess SF - Developing my Stockfish derivative

Post by AlexChess »

THE FUTURE:

Image

- 2xNNUE Linrock architecture
- Featuring 2 ProteusSFX-books (white and black) with all won games on Playchess.com and Lichess.org
- Supports Polyglot .bin and Chessbase .ctg books
- 20% faster than ProteusSFX-AI
- Based on the very latest Stockfish-dev 16.1 builds and all the official patches
- It will be released under GPLv3 when ready (+15 ELO than latest Stockfish build)

THE PRESENT:

Image
Is the eXperimental + AI module version with Learning

I'M VERY INTERESTED TO YOUR ANALYSIS AND EVALUATIONS
BUT PLEASE ADD THEM ON THE CONSECUTIVE DEDICATED POST
PROTEUS CHESS SF - COMMENTS


January, 13 2024 Update: First tests of ProteusSFX-AI are very good: Reached 3342 ELO Blitz on Playchess.com!

Image

Image

I thanks a lot Pham, Sarona and Mortal_sin for their support.

I'm returned to LOCAL TESTS, after 1 year of boring Playshit and Lichess.org (both fulls of hassholes) and I'm PRIVATELY developing ProteusSF-Piranha AI ONLY for myself.

January 13, 2024 update
As my friends here could have noticed, lately I'm moving to LOCAL engines tests.

I'm using Banksiagui 0.58 Windows to run all engines on the same hardware: 2014 Core i5 4 cores (2 MN/s)

As suggested by 2 dear chess friends, servers are only a waste of time since results are unreliables.

You can simply compare A2-a3 , Detlef Uter 32 or Eman cluster 40 threads Playchess users results against my Core I5 performances.

After 6000 games against most of the engines running on Playchess using this time the same hardware, I'm proud about ProteusSFX-AI! :mrgreen:

Image

Confirmed by 1000 games played against all best ORIGINAL engines, without losing a single game!

Image


November 23, 2023 update
Lichess.org results of ProteusSFX-AI and all bots (updated daily)

BULLET
BLITZ
RAPID
CLASSICAL

Top 50 Janualy 13, 2024 First Bullet, Blitz, Rapid & Classical

Image

Image
November 16. 2023 update Very good performance of ProteusSFX-AI!
First on Playchess.com right now with 3366 ELO blitz, 2499 bullet and 2529 long time!

Image

Image

Image
Is based on last Stockfish-dev build + AI module


After 3 months of developing in C++ I'm starting my first tests of ProteusSF-Piranha AI !!!

First implementation:

A supervisor asks to Bard and ChatGPT APIs the best strategy according to the opening when out of the book, then ProteusSF applies some pre-configuted patterns. Eg It doesn't try to win against C67 with White.

No learning on Piranha, only contestual analysis in real time, so the engine is even faster!

Like some leading chess programs, such as AlphaZero and Leela Chess Zero, that already use LLM to improve their performance.

LLM can be used to improve chess programs in a number of ways. First, LLM can be used to generate new game ideas. This can be done by using LLM to analyze a wide range of chess positions and identify possible moves that could lead to an advantage.

Second, LLM can be used to evaluate the strength of a chess position. This can be done by using LLM to analyze the possible moves of both sides and determine which position is more advantageous.

Third, LLM can be used to learn from chess games played by humans. This can be done by using LLM to analyze the games and identify the strategies and tactics that were used to win.

Some of the specific ways in which LLM can be used to improve chess programs include:

Move generator: LLM can be used to generate new game ideas that might not have been considered by a traditional chess engine.
Position evaluation: LLM can be used to evaluate the strength of a chess position more accurately than a traditional chess engine can.
Learning from chess games: LLM can be used to learn from chess games played by humans more efficiently than a traditional chess engine can.
Chess programs that use LLM have been shown to be able to beat traditional chess programs. For example, AlphaZero beat Stockfish, which was considered the best chess engine in the world at the time, in a series of games.

It is likely that LLM will continue to improve the performance of chess programs. As LLM technology continues to develop, chess programs that use LLM will become even stronger.

Additional information:

LLM stands for "large language model". It is a type of artificial intelligence (AI) that is trained on a massive dataset of text and code. LLMs can be used for a variety of tasks, including generating text, translating languages, writing different kinds of creative content, and answering questions in an informative way.


THE PAST:
Image

UPDATE October 25, 2023

Inspired by the Golden number, enhancing Stockfish propotions harmony.

ProteusSF-Aureo based on last Stockfish-dev 16.x & net, Learning
and up to 2 Polyglot books support is ready for CLOSED INTERNAL
betatests by microsmeta.com community. ...Request to partecipe!!

ProteusSF-Aureo on Playchess.com

Image

BEST PERFORMANCE EVER ON PLAYCHESS.COM SERVER. REALLY IMPRESSIVE RESULTS BY ProteusSF-Aureo RUNNING ON Core i5 4 THREADS AGAINST RYZEN AND THREADRIPPER 128 THREADS BOOSTED ENGINES AHEAD AI 50.3, ZEUS 50.3, Eman 9.41, Dark SisTer 7.80, Dragon 3.2, Stockfish 16 & dev230903, Corchess 4, ShashChess 33.2, BrainLearn 25.2, Octopussy, LC0 0.30.0, Raid, Lutefish, Alea Acta est, Sun and Smile!

NOTE: I consider Playchess my relative personal benchmark USING ALWAYS 4 CORES.

ProteusSF-Aureo FINAL on Playchess has reached 3326 ELO Blitz, 2468 bullet and 2509 ELO on Long Time!
Many thanks also to my loyal betatesters Mortal_Sin 3247 Blitz and ABcom 3326 with better hardware than mine!!!


Image

I'm very bored of Playchess.com for at least four reasons:

1) Since a decade a bug never fixed doesn't allow to correctly resume a running game when server suddenly disconnects and you lose 17 points!
2) It's almost an abandonware. Only 50 engines playing (multiple accounts by the same 30 users) and all with the same strength.All games draw!
3) It's available only for Windows, so I cannot exploit the power of my main computer Mac mini M1 8 cores ..I've only 4 cores of a Core I5 MiniPC
4) Chat is almost unmoderated, there are cheaters like Psycotrance playing with his other account AssRock J33455M and true idiots like TRULLI_

EVEN the very last GUI 18 of Chessbase Dragon 3.2 just purchased from Chess House has still the same bugs!!! :oops:

So I'm also playing online on the FREE & open source Lichess.org, full of statistics and I can follow all the games LIVE from the web or smartphone!


All Top Playchess.com engines 4 CPUs 2 MN/s Blitz 5+0 offline tournament

Image

All the games played in PGN. All engines playing with the same hardware (4 CPUs 2 MN/s)



ProteusSF Blitz 5+0 local gautlets on Chessbase GUI


UPDATE August 26, 2023

Image

ProteusSF-Aureo against 20 best engines: Stockfish, derivatives and originals. Finally with the same CPU speed (all 4 CPUs excepted LC0 0.30 2 CPUs, Dragon 3.2 that locks only 3 CPUs even changing the settings and Chess System Tal 2.0 DEMO limited to 1 CPU by design. My engine has already won a lot of games. I'm using the most common Playchess.com time control (5+0) against almost all engines playing there. Only AI and Zeus are 27.0 and I don't have MZ's HypnoS. I will run this gantlet by night while on daytime I'll continue my personal challenge on Playchess, trying to reach 3300 ELO before the others. I'm using my ProteusSF-Book.ctg for ProteusSF-Aureo and Fritz 18.ctg for all the other engines.


Image

ProteusSF on Lichess.org

I'D LIKE TO PLAY AGAINST YOU ALSO ONLINE USING FREE LICHESS CLIENT ON BANKANKSIAGUI, TELL ME IF YOU ARE AVAILABLE!
My accounts of Lichess.org: ProteusSF-Turbo 6 MN/s, ProteusSF 2 MN/s, ProteusSF-Lite 300 kN/s & ProteusSF-Open that uses the hardware choosen according to opponent strenght...Challenge me!

On Lichess.org few TOP BOTs have 3000-3200 ELO, while Leader BOTs have 2850 -->2900 ELO, due to the fact that a won game gives +6 points, not +17 points like on Playchess. All ProteusSF accounts (excepted Godot that plays another challenge) have just reached 3000+ ELO!

UPDATE October 4, 2023
LICHESS.org TOP 50 BOTS Bullet, Blitz, Rapid and Classical

Image

UPDATE July 10, 2023
Thanks to @nike22 just added ProteusSF-Godot automatically waiting challenges using @Torom BotLi Python script to study and fix most BanksiaGUI Lichess client bugs with Pham (BSG often hangs so you lose games only by time!)


I'm there even on Macmini 2010 300 kN/s (!) (no need to have Windows like with commercial Playchess.com or his clone Infinitychess.com finally) but there are less bots of top strengh. We need Dark Sister, Blue Marlin and friends. We can build there our personal free Arena :D





Image

According to the website lichess.org, the current ELO and ranking of all TOP BOTS in different chess variants are as follows:

ALL TOP BOTs ranking (no games against humans )

JUST BACK TO MY BANKSIAGUY TOP ENGINES TOURNAMENTS USING THE SAME HARDWARE AND TCEC23 FINAL 50 JEROEN NOOMEN PGN BOOK SWITCHING SIDES ON EVERY POSITION FOR FAIR RESULTS. SEND ME LINKS OF YOUR UPDATED ENGINES TO PARTECIPATE!

Image




ProteusSF-Sunrise Gauntlet against only top engines 3300-3600 ELO
Updale August, 12 2023

Image

BLITZ 3+0 ProteusSF-Aurora 230716 vs Stockfish 16 dev, SF derivatives and best originals

ALL 140 GAMES

BLITZ 3 MINUTES + 2 SECONDS / ALL - TCEC 22 UNBALANCED OPENING BOOK (50 openings switching sides, 100 games for each of 38 engines)
TABLEBASES SYZYGY 6 MAN, CROSSTABLE & PGN WITH GAMES UPDATED DAILY HARDWARE Core i5 4 CPUs 3 MN/s 8 GBRAM 1 GB hashtables

Engines with Learning feature: ProteusSF-Sunrise-221025, DarkSister 4.0, Solista Attack v2.0, AI 27.0 , BrainLearn 20, Eman 8.40

Proteus SF-Piranha 221016 (same mods of Sunrise 221025)
has won TCEC 22 Blitz Gauntlet 2 points beyond best competitor!!


I'm estremely happy about ProteusSF-Sunrise-221025 and ProteusSF-Piranha-221016 performances on my gauntlet ProteusSF-Sunrise-221025 versus all top engines (GPLv3 compliants, privates and originals) . The last one has the same SF mods excluded the power consuming learning that isn't suitable for computers with weaker hardware and few RAM. TCEC 22 Jeroom Noomen select 50 positions final is the most important openings testing suite to consider because all engine developers know and respect the author and the challengers are much stronger than the Official IGCA 2022 World Chess Champions Power Fritz 18 (Ginkgo) and Dragon 3.1

Image

ALL 3800 GAMES of the gauntlet


THE PAST:
Image

Interesting challenge between friends in a fair tournament. If I need an update, I cannot play again all the 3800 games of the gauntlet-holder Proteus-Sunrise or the match will completely restart, but updating progressively all challengers is the same. All engines are tested against an opponent with constant strength, reversing sides for each position. I can't imagine better rules: the test is more meaningful than on Playchess.com using customized books that alter results.

NEW! ProteusSF-Sunrise Rapid chess gauntled versus Stockfish-dev and Stockfish derivatives.
Guest star: Dragon 3.1 Note: All engines updated re-playing games when availables.

Time controls: 10 minutes + 4 seconds / all on Windows 11 ARM 64 4 CPUs (2 MN/s)
Opening Book: UHO 1.10-1.19 8mvs.PGN unbalanced reversing sides for a fair test.
50 starting positions, 100 games for each engine Tablebases: Syzygy 6 man. Updated: every 24 hours

Image

All 440 games until now

ENGINES SPEED BENCHMARKS ON THIS GAUNTLET

Image


Updated to 221025. Good performance on regression test also with double search speed 6 Mn/s on Mac M1 8 CPUs!
Image
All 100 PGN games TCEC22 UNBALANCED Added new positional / materialistic settings on config.

ProteusSF-Sunrise-221025 vs Stockfish-221027 61 - 59
Switching sides on each opening


DOTS6 unbalanced PGN test suite match 2 hours + 30 secs / all on Android 8 CPUs 400 kN/s  

Image

ALL 120 GAMES UNTIL NOW  Results updated every 48-72 hours

ProteusSF-Sunrise-221025 Rapid 10 + 4 secs /all Mini-Gauntlet
21 top engines ~3500 ELO reversing color on each position - Tablebases Syzygy 6 man - 2 MN/s on Windows 11 ARM64 4CPUs 4 GB RAM

Image

ALL 400 PGN GAMES

MATCH AGAINST STOCKFISH-DEV 221016
Opening book 3moves_Balanced PGN TC 1 hour + 15 secs / all on Mac mini 2010 (400 kN/s like Blitz on a 4Mn/s PC) Tablebases Syzygy 5 man

Image

All 100 PGN games 50 positions played by both color (100 games)

MATCH AGAINST STOCKFISH-DEV 221003
Opening book SuperGM_4moves pgn balanced TC 1 hour + 15 secs / all on Android (400 kn/s like blitz 5 mins + 2 secs on a 4 Mn/s PC)

ALL 128 PGN games64 positions played by both color (128 games)

AS YOU CAN SEE, BALANCED OPENINGS BOOK MATCHES ARE VERY BORING, ALL DRAWS!


PETER OSTERLUNG'S HARD-TALKCHESS 128 POSITION REVISED TEST
400 Kn/s search speed (218 times slower than a Threadripper 3995wx so a fair time limit for solving is 60 minutes instead of 15 seconds)

Image
Image

File name : Peter.epd
Total test items : 128
Test for : best moves
Total engines : 1
Timer : 1/3600+1
Expand ply : 2
Elapsed : 48:35:18
Laps : 1
Total tests : 128
Total corrects : 77 (60%)
Ave correct elapse : 37:51
Status : completed



MARCH 30, 2023 NEW RECORD BY ProteusSF-Crystal!


PETER OSTERLUNG'S HARD-TALKCHESS 128 POSITION REVISED TEST
400 Kn/s search speed (218 times slower than a Threadripper 3995wx so a fair time limit for solving is 60 minutes instead of 15 seconds)

Correct/Total:
ProteusSF-Crystal 230329: 107/128 (80%)

Failed tests (hit *):
1. ProteusSF-Crystal 230329:
Carlsen, M._Caruana, F._HTC108-7 WM2018, 3, Van Essen, M._2004, #22._HTC108-48 (-9ZNr41) S, 16, Los, Andrii_Kubicki, Tadeusz_corr LIPEAD40 (PER), 28, ACT3, ??._Festung_S, 34, Neghina, M._GoldenerKofig, 2009._S, 41, Ree, H._Hort, V._Wijk 1986, 42, Krug, P._CCC, Paloma_S, 46, Sikorsky, Horst_Vinchev, Simeon_corr GER, Hiarcs, 49, ERET 093_Endspiel T&S&B vs T&T&L_Simkhovich 1923, 64, ERET 094_Damenopfer_Sirobaba-Chaika, CorrGame 2011, 65, Fortress, Study?_?_ACT3-51 S, 66, ACT4, Nr.1_?_ACT4-1, 67, Cyclic, ZZ?._?_ACT5-25 2, 68, Zinar, M._HHdbVI.0210037.002a., 09076?._ACT5-44 2, 76, Bernhardt, G._?_ACT5-58 2, 81, Wotawa, A._?_ACT4-66 1, 98, Kraemer, A._?_ACT3-49, 106, Bazlo, Yuri_10moves, less_2019 HHdbVI 1, 112, Yarmonov, I._Glarean_2020, 117, Krug, P.._Glarean, EngineCrackers2_2018, 120, Josten, G.._Glarean, EngineCrackers_Studie 2001, 124

Successful tests:
1. ProteusSF-Crystal 230329:
HTC108, 2._Corbit, Dann_Tactical Insanity Nr.1, 1, TacticalInsanityNr.9_CorbitD_HTC108-5 S, 2, ACT-Suite, Nr.12_Dorsz, OpenChess_HTC108-8, 4, van Foreest, J._L'Ami, E._HTC108-9 Zug 24...Sideline, 5, Konig?, CCC._Cornforth96?, CSS+Playchess_HTC108-11 Arasan, 6, PoorFish, CostalbaM_Borderline, Dual_HTC108-12, 7, LC0, CCC._SF, CSS._HTC108-15 S, 8, AlphaZero, HTC108-21._SF_Game9 2017, 9, Gashimov, V._Azariov, S._2011 HTC108-23, 10, CorbitD, CCC+Rybkaf_?_HTC108-25, 11, Gusia_Barlov_Schweiz 1980 HTC108-28, 12, Bondarenko, &Kuznezov_Glarean, PoorStockfish_HTC108-29, 13, Rudolph, W.E._?_Studie 1912 HTC108-30 S, 14, Gusev, Y._Auerbach, E.25._HTC108-42, 15, Neghina, M._Stronghold, 2003?._HTC108-49 S, 17, Tunik_Duzhakov_2011 HTC108-52, 18, Neghina, M._Chase, Derived_HTC108-54, 19, SolomonKing, ?CCC._2.g4, Hiarcs_HTC108-80 S, 20, Nicolic, E._Fischer, R._HTC108-94 S, 21, Morozevich, A._Vitiugov, N._Reggio 2011 HTC108-100, 22, M. Garcia - P. Krug, Studie 2015_Glarean_The Engine Crackers, 23, Zepler, E.E._Studie, 1928._?, 24, Formela, Michal_Studie, 2019._Polski Zwiazek Szachowy, 25, Stoofvlees_SF_TCEC 2019, 26, Lasker, Emanuel_Lipke, Paul_Nebenvariante 41.Ke4! MS, 27, PurePower_Exxon_Chessbase online, 29, Bauer, Robert_Zhak, Boris Mikhailovich_WC30/final, 30, Mihai Neghina_Desperation_White to move and draw S, 31, Tauber, Hans_Sikorsky, Horst_37.e5!, 32, Bondarenko+Gorgiev, Studie1959_BrillanterSchachzug, 17._?, 33, ACT3, ??._Studie?_?, 35, ACT3, ??._?_?, 36, ACT3, ??._Festung_?, 37, ACT3, ??._Festung_S, 38, ACT3, ??._Studie?_?, 39, ACT3, ??._Studie?_S, 40, Neghina, M._WhieTo, MoveAndWin_2018? S, 43, Popov, Study_Glarean 2006, 1996?._HTC108-58, 44, Sikorsky, Horst_Novak, Joze_Sideline, 45, TCEC, Sammlung_Nr.8, j329_CCC, 47, TCEC, Sammlung_Nr.22, rj329_CCC, 48, Zemljanski, Y.I._1move, cut_Tsjechover-90 S, 50, Study, ?._?_IQ 100, 51, Study?, ?._?_ACT3, 52, Study?, ?._?_TTT2 31, 53, Semenikov+, D A._Sikorsky, H._corr. USA 2021, 54, Petrov, M._Sikorsky, H._CL/2021, 55, Study, ?._CCC_ACT5 11, 56, Study, ?._?_ACT5 29, 57, ERET 015_Endspiel D&B vs T&L&B_Becker 2015, 58, ERET 032_Zugzwang_Fahrni 1922, 59, ERET 037_Zugzwang_Garcia 2000, 60, ERET 046_Endspiel T&S vs T&S_Grandelius-Raznikov, Albena 2011, 61, ERET 050_Koenigsangriff_Hansen-Barua, Biel 1993, 62, ERET 090_Endspiel L vs L_Salai 2011, 63, Blass, U.._CCC_ACT5-27 2, 69, Campioli, M._#14, ?._ACT5-28 2, 70, Krug, P._?_ACT5-30 2, 71, KrugP, Garcia_HHdbVI.0044383.001, a.01968?_ACT5-31 1, 72, Kos, V._?_ACT5-34 2, 73, Sokka, H._HHdbVI.1076075.001, a.53387?_ACT5-36 1, 74, Sehwers, J?._HHdbVI.1643584.002a., 89051?._ACT5-ana, 75, Nestorescu, V._HHdbVI.0708805.002a., 33432?._ACT5-46 1, 77, Timman, J._HHdbVI.0033988.002a., 01480._ACT5-48 2, 78, Unknown, Author_?_ACT5-51 2, 79, Ivanov, V._?_ACT5-53 2, 80, Unknows, Author_?_ACT5-61 2, 82, Naef, W._?_ACT5-63 2, 83, #7, ?._?_ACT5-67 2, 84, #9, ?._?_ACT5-68 2, 85, SilhanJ, ZizkaL_?_ACT5-59 3, 86, Hiarcs, ?._?_TTT2 2, 87, Hiarcs, ?._?_TTT2 6 2, 88, Derived, from_StavrietskyA, 2017._TTT2 14, 89, Midgame, ?._?._TTT2 23, 90, TTT2.053_?_TTT2, 91, TTT2.073_?_TTT2, 92, RoxlauM, GrushkoM_HHdbVI_ACT4 4 1, 93, Smyslov, V._?_ACT4-26 1, 94, Smyslov, V._?_ACT4-41 1, 95, JostenG_?_?, 96, Study, ?._?_ACT4-59, 97, Study, ?._?_ACT4-90 1, 99, Bethge, F._?_ACT4-91, 100, Blathy, O._?_ACT4-94 1, 101, Cortlever, N._HHdbVI_ACT3-29 1, 102, Korolkov, V._?_ACT3-38, 103, Fortress, ?._?_ACT3-41 1, 104, Unknown, Author_?_ACT3-48, 105, Sivkov, N._#10, ?._ACT3-88, 107, Timman=J_Glarean, Nightmare_?, 108, Boudy Bueno, J.L.._Szabo, L._Varna 1979 Txb2!? Eret 17, 109, Lc0-dev_Stockfish_ECO MegaMatch II! (1|1), 110, Carlsen, M._L'Ami, E._Corus B VLH16 35, 111, Opening, Hiarcs_!, ?!._?, 113, TCEC, rj329_?_CCC, 114, TCEC, rj329_?_CCC Nr.5, 115, Gleichmann, Matthias_Praznik, Niko_corr CT21/final, 116, Eman_SugaR_Glarean, 118, Hobacle's, VersionOf_Walter Eigenmann's, Study_?, 119, Gniazdowski, Marian_Rau, Johann_corr CL/2019/C8 Glarean, 121, Illescas, Miguel_Glarean, EngineCrackers_?, 122, Heinz, J._Glarean_Study 1981, 123, Timman, J.._HHdbVI_?, 125, Gelly, M.._HHdbVI_Glarean, 126, Krug, P._HHdbVI, Glarean_?, 127, Ganguly, S._RuszA, correction_HHdbVI, 128


Correct/Total:
ProteusSF-Sunrise 221015: 77/128

Image

Failed tests (hit *):
1. ProteusSF-Sunrise 221015:
TacticalInsanityNr.9_CorbitD_HTC108-5 S, 2, Carlsen, M._Caruana, F._HTC108-7 WM2018, 3, Gashimov, V._Azariov, S._2011 HTC108-23, 10, Gusia_Barlov_Schweiz 1980 HTC108-28, 12, Bondarenko, &Kuznezov_Glarean, PoorStockfish_HTC108-29, 13, Rudolph, W.E._?_Studie 1912 HTC108-30 S, 14, Van Essen, M._2004, #22._HTC108-48 (-9ZNr41) S, 16, Neghina, M._Chase, Derived_HTC108-54, 19, Formela, Michal_Studie, 2019._Polski Zwiazek Szachowy, 25, ACT3, ??._Studie?_?, 35, ACT3, ??._?_?, 36, ACT3, ??._Festung_?, 37, ACT3, ??._Festung_S, 38, ACT3, ??._Studie?_?, 39, Neghina, M._WhieTo, MoveAndWin_2018? S, 43, Popov, Study_Glarean 2006, 1996?._HTC108-58, 44, Sikorsky, Horst_Novak, Joze_Sideline, 45, Krug, P._CCC, Paloma_S, 46, Sikorsky, Horst_Vinchev, Simeon_corr GER, Hiarcs, 49, Zemljanski, Y.I._1move, cut_Tsjechover-90 S, 50, Study, ?._?_IQ 100, 51, Semenikov+, D A._Sikorsky, H._corr. USA 2021, 54, Study, ?._?_ACT5 29, 57, ERET 032_Zugzwang_Fahrni 1922, 59, ERET 090_Endspiel L vs L_Salai 2011, 63, ERET 093_Endspiel T&S&B vs T&T&L_Simkhovich 1923, 64, ERET 094_Damenopfer_Sirobaba-Chaika, CorrGame 2011, 65, Fortress, Study?_?_ACT3-51 S, 66, Cyclic, ZZ?._?_ACT5-25 2, 68, Zinar, M._HHdbVI.0210037.002a., 09076?._ACT5-44 2, 76, Bernhardt, G._?_ACT5-58 2, 81, Derived, from_StavrietskyA, 2017._TTT2 14, 89, TTT2.053_?_TTT2, 91, Smyslov, V._?_ACT4-26 1, 94, Smyslov, V._?_ACT4-41 1, 95, Study, ?._?_ACT4-59, 97, Wotawa, A._?_ACT4-66 1, 98, Blathy, O._?_ACT4-94 1, 101, Cortlever, N._HHdbVI_ACT3-29 1, 102, Korolkov, V._?_ACT3-38, 103, Unknown, Author_?_ACT3-48, 105, Kraemer, A._?_ACT3-49, 106, Bazlo, Yuri_10moves, less_2019 HHdbVI 1, 112, TCEC, rj329_?_CCC Nr.5, 115, Yarmonov, I._Glarean_2020, 117, Eman_SugaR_Glarean, 118, Hobacle's, VersionOf_Walter Eigenmann's, Study_?, 119, Krug, P.._Glarean, EngineCrackers2_2018, 120, Illescas, Miguel_Glarean, EngineCrackers_?, 122, Gelly, M.._HHdbVI_Glarean, 126, Ganguly, S._RuszA, correction_HHdbVI, 128

Successful tests:
1. ProteusSF-Sunrise 221015:
HTC108, 2._Corbit, Dann_Tactical Insanity Nr.1, 1, ACT-Suite, Nr.12_Dorsz, OpenChess_HTC108-8, 4, van Foreest, J._L'Ami, E._HTC108-9 Zug 24...Sideline, 5, Konig?, CCC._Cornforth96?, CSS+Playchess_HTC108-11 Arasan, 6, PoorFish, CostalbaM_Borderline, Dual_HTC108-12, 7, LC0, CCC._SF, CSS._HTC108-15 S, 8, AlphaZero, HTC108-21._SF_Game9 2017, 9, CorbitD, CCC+Rybkaf_?_HTC108-25, 11, Gusev, Y._Auerbach, E.25._HTC108-42, 15, Neghina, M._Stronghold, 2003?._HTC108-49 S, 17, Tunik_Duzhakov_2011 HTC108-52, 18, SolomonKing, ?CCC._2.g4, Hiarcs_HTC108-80 S, 20, Nicolic, E._Fischer, R._HTC108-94 S, 21, Morozevich, A._Vitiugov, N._Reggio 2011 HTC108-100, 22, M. Garcia - P. Krug, Studie 2015_Glarean_The Engine Crackers, 23, Zepler, E.E._Studie, 1928._?, 24, Stoofvlees_SF_TCEC 2019, 26, Lasker, Emanuel_Lipke, Paul_Nebenvariante 41.Ke4! MS, 27, Los, Andrii_Kubicki, Tadeusz_corr LIPEAD40 (PER), 28, PurePower_Exxon_Chessbase online, 29, Bauer, Robert_Zhak, Boris Mikhailovich_WC30/final, 30, Mihai Neghina_Desperation_White to move and draw S, 31, Tauber, Hans_Sikorsky, Horst_37.e5!, 32, Bondarenko+Gorgiev, Studie1959_BrillanterSchachzug, 17._?, 33, ACT3, ??._Festung_S, 34, ACT3, ??._Studie?_S, 40, Neghina, M._GoldenerKofig, 2009._S, 41, Ree, H._Hort, V._Wijk 1986, 42, TCEC, Sammlung_Nr.8, j329_CCC, 47, TCEC, Sammlung_Nr.22, rj329_CCC, 48, Study?, ?._?_ACT3, 52, Study?, ?._?_TTT2 31, 53, Petrov, M._Sikorsky, H._CL/2021, 55, Study, ?._CCC_ACT5 11, 56, ERET 015_Endspiel D&B vs T&L&B_Becker 2015, 58, ERET 037_Zugzwang_Garcia 2000, 60, ERET 046_Endspiel T&S vs T&S_Grandelius-Raznikov, Albena 2011, 61, ERET 050_Koenigsangriff_Hansen-Barua, Biel 1993, 62, ACT4, Nr.1_?_ACT4-1, 67, Blass, U.._CCC_ACT5-27 2, 69, Campioli, M._#14, ?._ACT5-28 2, 70, Krug, P._?_ACT5-30 2, 71, KrugP, Garcia_HHdbVI.0044383.001, a.01968?_ACT5-31 1, 72, Kos, V._?_ACT5-34 2, 73, Sokka, H._HHdbVI.1076075.001, a.53387?_ACT5-36 1, 74, Sehwers, J?._HHdbVI.1643584.002a., 89051?._ACT5-ana, 75, Nestorescu, V._HHdbVI.0708805.002a., 33432?._ACT5-46 1, 77, Timman, J._HHdbVI.0033988.002a., 01480._ACT5-48 2, 78, Unknown, Author_?_ACT5-51 2, 79, Ivanov, V._?_ACT5-53 2, 80, Unknows, Author_?_ACT5-61 2, 82, Naef, W._?_ACT5-63 2, 83, #7, ?._?_ACT5-67 2, 84, #9, ?._?_ACT5-68 2, 85, SilhanJ, ZizkaL_?_ACT5-59 3, 86, Hiarcs, ?._?_TTT2 2, 87, Hiarcs, ?._?_TTT2 6 2, 88, Midgame, ?._?._TTT2 23, 90, TTT2.073_?_TTT2, 92, RoxlauM, GrushkoM_HHdbVI_ACT4 4 1, 93, JostenG_?_?, 96, Study, ?._?_ACT4-90 1, 99, Bethge, F._?_ACT4-91, 100, Fortress, ?._?_ACT3-41 1, 104, Sivkov, N._#10, ?._ACT3-88, 107, Timman=J_Glarean, Nightmare_?, 108, Boudy Bueno, J.L.._Szabo, L._Varna 1979 Txb2!? Eret 17, 109, Lc0-dev_Stockfish_ECO MegaMatch II! (1|1), 110, Carlsen, M._L'Ami, E._Corus B VLH16 35, 111, Opening, Hiarcs_!, ?!._?, 113, TCEC, rj329_?_CCC, 114, Gleichmann, Matthias_Praznik, Niko_corr CT21/final, 116, Gniazdowski, Marian_Rau, Johann_corr CL/2019/C8 Glarean, 121, Heinz, J._Glarean_Study 1981, 123, Josten, G.._Glarean, EngineCrackers_Studie 2001, 124, Timman, J.._HHdbVI_?, 125, Krug, P._HHdbVI, Glarean_?, 127


THE PRESENT:

GPLv3 compliant ProteusSF-Piranha 220904 is available to the public for WIndows, Mac and Android




Faster, without experience feature and directly based on Stockfish latest source (no more from other derivatives)

Image


Testing it at longer time controls, with and without NNUE, with tablebases Syzygy 6 man and balanced 2 moves book.
Polyglot bin book support is available after a lot of requests, but like @dorsz I didn't like customized .bin books that
alter tests results. I prefer PGN test suites switching colors for a fair match. For my comparatives I will use only latest
Stockfish-dev
and top original engines, no more weak random engines as done by Chess Engines Diary and other testers.
Authors of original free open source engines like Berserk 9 and Koivisto 8.1.3, so close in strength to Stockfish, are
much more skilled than me and I have a lot to learn studying their smart code :oops:

CCRL and CEGT rating lists are playing useless blitz of thousand of unreliable games! Obsolete Stockfish 15 220418
3592 ELO with only 1CPU on old PCs? The same engine is rated 3744 ELO 2 min 1s /all on a 2013 Core I7 4770k ? :shock:

SPCC Rating list is reliable and updated, but it runs 20 concurrent games at 750 kN/s and it overestimates the strength by 300 ELO!

IPMAN CHESS Rating list calculated on a Intel Core i9 7980XE @4.2Ghz with TC 10s + 1 sec is by far my preferred!

For me Stockfish and all derivatives are 3500 Elo blitz and 3300 ELO on long time controls, using 2x-3x faster hardware!

Image

TESTING GOLDEN RULES
-Avoid tests made using opening books without switching colors for every position, especially if using unbalanced openings.
-Prefer PGN test suites like unbalanced DOTS5 or TCEC22 superfinal or neutral +-0.30 score books, swapping sides for each position.
-Always set ponder OFF. This options uses opponent's CPU cycles and is suited only for human vs engine matches.
-Disable experience files with weak hardware. They incredibly slow down the engine so there is no advantage but penalty.
-Don't use 2 GB hashtables on blitz. They will never been used and sometimes they crash the system due to memory lack

Image

ProteusSF-Crystal 221027 Peter Osterlung's 128 HardTalkchess positions revised 75%!
400 Kn/s search speed (218 times slower than a Threadripper 3995wx so a fair time limit for solving is 60 minutes instead of 15 seconds)



Correct/Total:
ProteusSF-Crystal 5 KWK 96/128

Image

File name : Peter.epd
Total test items : 128
Test for : best moves
Total engines : 1
Timer : 1/3600+1
Expand ply : 2
Elapsed : 39:18:47
Laps : 1
Total tests : 128
Total corrects : 96 (75%)
Ave correct elapse : 24:49
Status : completed


LAP 1:


Correct/Total:
ProteusSF-Crystal 5 KWK: 96/128

Failed tests (hit *):
1. ProteusSF-Crystal 5 KWK:
Carlsen, M._Caruana, F._HTC108-7 WM2018, 3, Los, Andrii_Kubicki, Tadeusz_corr LIPEAD40 (PER), 28, ACT3, ??._Festung_S, 34, ACT3, ??._Festung_?, 37, ACT3, ??._Festung_S, 38, ACT3, ??._Studie?_?, 39, ACT3, ??._Studie?_S, 40, Sikorsky, Horst_Novak, Joze_Sideline, 45, Sikorsky, Horst_Vinchev, Simeon_corr GER, Hiarcs, 49, Study, ?._CCC_ACT5 11, 56, ERET 093_Endspiel T&S&B vs T&T&L_Simkhovich 1923, 64, ERET 094_Damenopfer_Sirobaba-Chaika, CorrGame 2011, 65, Fortress, Study?_?_ACT3-51 S, 66, ACT4, Nr.1_?_ACT4-1, 67, Cyclic, ZZ?._?_ACT5-25 2, 68, Zinar, M._HHdbVI.0210037.002a., 09076?._ACT5-44 2, 76, Bernhardt, G._?_ACT5-58 2, 81, SilhanJ, ZizkaL_?_ACT5-59 3, 86, Smyslov, V._?_ACT4-26 1, 94, Smyslov, V._?_ACT4-41 1, 95, Wotawa, A._?_ACT4-66 1, 98, Blathy, O._?_ACT4-94 1, 101, Cortlever, N._HHdbVI_ACT3-29 1, 102, Unknown, Author_?_ACT3-48, 105, Kraemer, A._?_ACT3-49, 106, Bazlo, Yuri_10moves, less_2019 HHdbVI 1, 112, TCEC, rj329_?_CCC Nr.5, 115, Krug, P.._Glarean, EngineCrackers2_2018, 120, Gniazdowski, Marian_Rau, Johann_corr CL/2019/C8 Glarean, 121, Illescas, Miguel_Glarean, EngineCrackers_?, 122, Heinz, J._Glarean_Study 1981, 123, Gelly, M.._HHdbVI_Glarean, 126

Successful tests:
1. ProteusSF-Crystal 5 KWK:
HTC108, 2._Corbit, Dann_Tactical Insanity Nr.1, 1, TacticalInsanityNr.9_CorbitD_HTC108-5 S, 2, ACT-Suite, Nr.12_Dorsz, OpenChess_HTC108-8, 4, van Foreest, J._L'Ami, E._HTC108-9 Zug 24...Sideline, 5, Konig?, CCC._Cornforth96?, CSS+Playchess_HTC108-11 Arasan, 6, PoorFish, CostalbaM_Borderline, Dual_HTC108-12, 7, LC0, CCC._SF, CSS._HTC108-15 S, 8, AlphaZero, HTC108-21._SF_Game9 2017, 9, Gashimov, V._Azariov, S._2011 HTC108-23, 10, CorbitD, CCC+Rybkaf_?_HTC108-25, 11, Gusia_Barlov_Schweiz 1980 HTC108-28, 12, Bondarenko, &Kuznezov_Glarean, PoorStockfish_HTC108-29, 13, Rudolph, W.E._?_Studie 1912 HTC108-30 S, 14, Gusev, Y._Auerbach, E.25._HTC108-42, 15, Van Essen, M._2004, #22._HTC108-48 (-9ZNr41) S, 16, Neghina, M._Stronghold, 2003?._HTC108-49 S, 17, Tunik_Duzhakov_2011 HTC108-52, 18, Neghina, M._Chase, Derived_HTC108-54, 19, SolomonKing, ?CCC._2.g4, Hiarcs_HTC108-80 S, 20, Nicolic, E._Fischer, R._HTC108-94 S, 21, Morozevich, A._Vitiugov, N._Reggio 2011 HTC108-100, 22, M. Garcia - P. Krug, Studie 2015_Glarean_The Engine Crackers, 23, Zepler, E.E._Studie, 1928._?, 24, Formela, Michal_Studie, 2019._Polski Zwiazek Szachowy, 25, Stoofvlees_SF_TCEC 2019, 26, Lasker, Emanuel_Lipke, Paul_Nebenvariante 41.Ke4! MS, 27, PurePower_Exxon_Chessbase online, 29, Bauer, Robert_Zhak, Boris Mikhailovich_WC30/final, 30, Mihai Neghina_Desperation_White to move and draw S, 31, Tauber, Hans_Sikorsky, Horst_37.e5!, 32, Bondarenko+Gorgiev, Studie1959_BrillanterSchachzug, 17._?, 33, ACT3, ??._Studie?_?, 35, ACT3, ??._?_?, 36, Neghina, M._GoldenerKofig, 2009._S, 41, Ree, H._Hort, V._Wijk 1986, 42, Neghina, M._WhieTo, MoveAndWin_2018? S, 43, Popov, Study_Glarean 2006, 1996?._HTC108-58, 44, Krug, P._CCC, Paloma_S, 46, TCEC, Sammlung_Nr.8, j329_CCC, 47, TCEC, Sammlung_Nr.22, rj329_CCC, 48, Zemljanski, Y.I._1move, cut_Tsjechover-90 S, 50, Study, ?._?_IQ 100, 51, Study?, ?._?_ACT3, 52, Study?, ?._?_TTT2 31, 53, Semenikov+, D A._Sikorsky, H._corr. USA 2021, 54, Petrov, M._Sikorsky, H._CL/2021, 55, Study, ?._?_ACT5 29, 57, ERET 015_Endspiel D&B vs T&L&B_Becker 2015, 58, ERET 032_Zugzwang_Fahrni 1922, 59, ERET 037_Zugzwang_Garcia 2000, 60, ERET 046_Endspiel T&S vs T&S_Grandelius-Raznikov, Albena 2011, 61, ERET 050_Koenigsangriff_Hansen-Barua, Biel 1993, 62, ERET 090_Endspiel L vs L_Salai 2011, 63, Blass, U.._CCC_ACT5-27 2, 69, Campioli, M._#14, ?._ACT5-28 2, 70, Krug, P._?_ACT5-30 2, 71, KrugP, Garcia_HHdbVI.0044383.001, a.01968?_ACT5-31 1, 72, Kos, V._?_ACT5-34 2, 73, Sokka, H._HHdbVI.1076075.001, a.53387?_ACT5-36 1, 74, Sehwers, J?._HHdbVI.1643584.002a., 89051?._ACT5-ana, 75, Nestorescu, V._HHdbVI.0708805.002a., 33432?._ACT5-46 1, 77, Timman, J._HHdbVI.0033988.002a., 01480._ACT5-48 2, 78, Unknown, Author_?_ACT5-51 2, 79, Ivanov, V._?_ACT5-53 2, 80, Unknows, Author_?_ACT5-61 2, 82, Naef, W._?_ACT5-63 2, 83, #7, ?._?_ACT5-67 2, 84, #9, ?._?_ACT5-68 2, 85, Hiarcs, ?._?_TTT2 2, 87, Hiarcs, ?._?_TTT2 6 2, 88, Derived, from_StavrietskyA, 2017._TTT2 14, 89, Midgame, ?._?._TTT2 23, 90, TTT2.053_?_TTT2, 91, TTT2.073_?_TTT2, 92, RoxlauM, GrushkoM_HHdbVI_ACT4 4 1, 93, JostenG_?_?, 96, Study, ?._?_ACT4-59, 97, Study, ?._?_ACT4-90 1, 99, Bethge, F._?_ACT4-91, 100, Korolkov, V._?_ACT3-38, 103, Fortress, ?._?_ACT3-41 1, 104, Sivkov, N._#10, ?._ACT3-88, 107, Timman=J_Glarean, Nightmare_?, 108, Boudy Bueno, J.L.._Szabo, L._Varna 1979 Txb2!? Eret 17, 109, Lc0-dev_Stockfish_ECO MegaMatch II! (1|1), 110, Carlsen, M._L'Ami, E._Corus B VLH16 35, 111, Opening, Hiarcs_!, ?!._?, 113, TCEC, rj329_?_CCC, 114, Gleichmann, Matthias_Praznik, Niko_corr CT21/final, 116, Yarmonov, I._Glarean_2020, 117, Eman_SugaR_Glarean, 118, Hobacle's, VersionOf_Walter Eigenmann's, Study_?, 119, Josten, G.._Glarean, EngineCrackers_Studie 2001, 124, Timman, J.._HHdbVI_?, 125, Krug, P._HHdbVI, Glarean_?, 127, Ganguly, S._RuszA, correction_HHdbVI, 128



Image

My other experimental tactical engine with NNUE but not based on Stockfish or derivatives

I'M VERY INTERESTED TO YOUR ANALYSIS AND EVALUATIONS
BUT PLEASE ADD THEM ON THE CONSECUTIVE DEDICATED POST
PROTEUS CHESS SF - COMMENTS



ProteusSF development begins on Jan, 2022

Image

Hi chess engine fans!

On this topic we discuss about how to modify Stockfish to create a DERIVATIVE engine focused on material sacrifices for positional advantage.

All ideas, suggestions and comments are welcome!

Image

I will release ProteusChessSF under GPLv3 attributing the original source to the Stockfish Team and never share my private engine executable before it will be really different in standard SF playing style.

At this point, even if the NNUE and MCTS support is available, I'm not interested to the strength but to the style of the engine (I want add positional sacrifices, king safety and castle assault patterns) so hand-crafted evaluation is useful :) Then I will add hybrid mode.


Image

First heavy modifications in Pawn Bonus values seem interesting, replaced SugaR 2.50 ICCF (closed project) with ProteusChessSF on my LT tournament :)

Image

Wow! Disabling NNUE (to use hand-crafted evaluations) Proteus is 3 times faster than the other SF derivative Kayra 1.1

Until now, my bonus pawns modifications to enforce pawn structure seems to compensate well the lack of a neural network. I will publish all the games In PGN draws the first 2 games against Fat Titz 2 130121 and Kayra 1.1 :D )


Image

Just challenging my dear friend Mehmet's very strong creature with my ONE MORE Stockfish derivative :lol:

A draw and a very nice win of Kayra 1.1 (actually #1 on my Group A LT tournament) against Proteus Chess SF v002 (classic evaluation, no NNUE) . [EDIT] Another interesting game lost by ProteusChess v002 against CorChess 3 100122 despite an advantage after opening (complex position) I will publish all LT tournament Proteus Chess SF games.

You can find the crosstable and all the PGN games here:

https://www.chess.com/forum/view/game-a ... -17?page=1
Locked