Chess2u
Would you like to react to this message? Create an account in a few clicks or log in to continue.

Chess2uLog in

Proteus Chess SF v004

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
AlexChess wrote:
deeds wrote:
For whose in denial of reality :

https://www.chess2u.com/t18033p25-proteus-chess-sf-v004#114181

An engine that learns it's smarter than an engine that searches in vain.

A book that improves the WDL ratio it's smarter than tens of thousands games to reduce the error margin.

Tablebases that solve a fraction of chess it's smarter than engines that analyze the same positions over and over again.



They play ten thousand games to reduce the error margin on ELO calculation, not to improve the WDL ratio.

But I prefer to play a fraction of games at a decent depth using all 8 CPUs instead of playing 8 games using only 1 CPU simply to say "I have played 4000 games (full of blunders) . I'm very professional"
lol

LOL..Don't embarrass yourself anymore, stop saying bullshit.. You have 0 idea about what you are saying..
It will be much better if you tell everyone what diagnosis the psychiatrist has given you.. LOL..

Although no, i changed my mind, don't stop talking, we want to laugh even more. lol lol lol lol lol

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
They've had absolute faith in the Fishtest sect for so many years that it scares them that there are other more effective, more advanced, more intelligent solutions. And when they are forced to recognize that tablebases or NNUE technology bring a "ten years" ELO gain, they add them to their archaic system so as not to seem dumber.

And all the while, their archaic system still hasn't solved any tiny fraction of the chess, they're analyzing positions in the void over and over again. They don't even store the positions with mate in X moves, no they prefer to reanalyze them again and again.

It took months for them to integrate the tablebases into SF, it took A0 to put them to shame for them to accept against their will that a gpu engine was a smart solution (Lc0), it still took months so that they integrate the NNUE technology, and since they saturate the size of their network every 2 months, they remove patches then put them back.

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
AlexChess wrote:
They play ten thousand games to reduce the error margin on ELO calculation, not to improve the WDL ratio.

But I prefer to play a fraction of games at a decent depth using all 8 CPUs instead of playing 8 games using only 1 CPU simply to say "I have played 4000 games (full of blunders) . I'm very professional" lol

Private engines learn, Stockfish can't learn, tens thousands games later :
Proteus Chess SF v004 - Page 3 2yn4fim

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
deeds wrote:
AlexChess wrote:
They play ten thousand games to reduce the error margin on ELO calculation, not to improve the WDL ratio.

But I prefer to play a fraction of games at a decent depth using all 8 CPUs instead of playing 8 games using only 1 CPU simply to say "I have played 4000 games (full of blunders) . I'm very professional" lol

Private engines learn, Stockfish can't learn, tens thousands games later :
Proteus Chess SF v004 - Page 3 2yn4fim


And what about sharing learned data on a server with access to all the instances of an engine?

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
The learned data is attached to the engine release. The engine release is optimized with a net. The same net doesn't bring the same elo with different engines. The experience data contains engine's signature for each position, engine's evaluation for each move, etc.

No interest to use a stockfish nnue net with dragon. So no interest to rent as many server as private engines with learning feature.

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
deeds wrote:
The learned data is attached to the engine release. The engine release is optimized with a net. The same net doesn't bring the same elo with different engines. The experience data contains engine's signature for each position, engine's evaluation for each move, etc.

No interest to use a stockfish nnue net with dragon. So no interest to rent as many server as private engines with learning feature.


Ok but imagine if the learning file could be updated instantly when an engine wins or loses sharing It between all the instances running at the same time...

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
AlexChess wrote:
Ok but imagine if the learning file could be updated instantly when an engine wins or loses sharing It between all the instances running at the same time...

Problem #1 : several concurrent accesses to the same file = collision, overwriting, blocking

Problem #2 : hundreds MB "updated instantly" over internet !? You dream !  In addition it means that you have to download the file between each game, no way

Problem #3 : if you lose against stronger opponents, do you decrease the move's quality in the same way as if you lose on time or by tablebases or by connection stalls or by a too longer loading time ? Cloud isn't reliable.

Problem #4 : if you win because your hardware is better, nothing prove your moves were not efficient on long term.

Problem #5 : what happend when the author won't update the engine ? Learned data => evaluation feature => engine release

Problem #6 : what happend when the filesize of the learned data weighs tens GB ?

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
deeds wrote:
AlexChess wrote:
Ok but imagine if the learning file could be updated instantly when an engine wins or loses sharing It between all the instances running at the same time...

Problem #1 : several concurrent accesses to the same file = collision, overwriting, blocking

Problem #2 : hundreds MB "updated instantly" over internet !? You dream !  In addition it means that you have to download the file between each game, no way

Problem #3 : if you lose against stronger opponents, do you decrease the move's quality in the same way as if you lose on time or by tablebases or by connection stalls or by a too longer loading time ? Cloud isn't reliable.

Problem #4 : if you win because your hardware is better, nothing prove your moves were not efficient on long term.

Problem #5 :  what happend when the author won't update the engine ? Learned data => evaluation feature => engine release

Problem #6 : what happend when the filesize of the learned data weighs tens GB ?


All true, but even NNUE was a dream only 5 years ago. I'm brainstorming about it with very skilled programmers. Thank you for you constructive answer smile

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
Proteus Chess SF v004 - Page 3 Zeus_210


Thanks to @Oliprog and all my friends. Mehmet I have compiled your Kayra 1.1 also for macOS M1, because I'm moving to macOS my tournaments to have a better hardware for ProteusChessSFv005 tests and I need the stronger engines to challenge.

Respect for Zeus Team, really sorry that you do not share this wonderful engine, I still use Zeus 11 engine and his net privately smile

Best regards, Alex

descriptionProteus Chess SF v004 - Page 3 EmptyRe: Proteus Chess SF v004

more_horiz
Permissions in this forum:
You cannot reply to topics in this forum