personal experience regarding "self learning"

[deeds]

How can the use of experience files/engines/books be considered cheating insofar as OTB play is concerned?

All the technologies which can beat sf will be firstly considered as cheating devices, unfair conditions, sss, etc until they will be added to sf (e.g. lazysmp, syzygy, nnue, etc.).

I suppose it's the same behavior between players on the board when the highest elo didn't won.

[rulleeeee]

https://drive.google.com/file/d/17OT2xq ... share_link

Hey deeds. I wanted to share with my experience file. You share your knowledge and programs, so I want to repay you by sharing my experience file with you. It has never been combined with other files.

I hope that my file will somehow help your project with Eman

Games were played from scratch and lasted about 1 hour/game

[deeds]

Hey deeds. I wanted to share with my experience file. You share your knowledge and programs, so I want to repay you by sharing my experience file with you.

Thanks Rulleeeee, since I opened the sales of my experience file, every day I measure the gratitude for sharing my tools and my knowledge.

It has never been combined with other files.

I hope that my file will somehow help your project with Eman

Sorry but you understand that to use your experience file, I would have to merge it with mine so I could no longer guarantee mine.
On the other hand, if your experience file was trained on openings where mine has few experience data, i may add these openings into my next learning sessions.

Games were played from scratch and lasted about 1 hour/game

With this kind of TC, the experience data are few (=low efficiency) but they are of good quality (=accurate scores).
One does not go without the other, you need both otherwise the engine does not learn.
And pay attention to the imbalance of your trained ECO codes...

[rulleeeee]

I understand

Eman himself chose the openings he wants to play and that's probably why there are no other openings.

Maybe because he found them weak.

If it wasn't a problem, could you do a little test to run your trained file vs mine?

I'd like to know how far mine is from yours

[deeds]

If it wasn't a problem, could you do a little test to run your trained file vs mine ?

What TC ?
What openings ?
How many games ?
Do I have to film the tournament ?
Do the games have to be public ?

[rulleeeee]

TC = 1+1
Berlin defence = r1bqkb1r/pppp1ppp/2n2n2/1B2p3/4P3/5N2/PPPP1PPP/RNBQK2R w KQkq - 0 4

Gucio Piano = r1bqkbnr/pppp1ppp/2n5/4p3/2B1P3/5N2/PPPP1PPP/RNBQK2R b KQkq - 0 3

Sicilian Defence = rnbqkbnr/pp1ppppp/8/2p5/4P3/5N2/PPPP1PPP/RNBQKB1R b KQkq - 0 2

1,000 opening games (500 white and 5 black)

Don't film games

Games and results can be shared

[deeds]

TC = 1+1
Berlin defence = r1bqkb1r/pppp1ppp/2n2n2/1B2p3/4P3/5N2/PPPP1PPP/RNBQK2R w KQkq - 0 4

Gucio Piano = r1bqkbnr/pppp1ppp/2n5/4p3/2B1P3/5N2/PPPP1PPP/RNBQK2R b KQkq - 0 3

Sicilian Defence = rnbqkbnr/pp1ppppp/8/2p5/4P3/5N2/PPPP1PPP/RNBQKB1R b KQkq - 0 2

1,000 opening games (500 white and 5 black)

Don't film games

Games and results can be shared

TC 1m+1s => 3 min/game
1000 games * 3 openings => 3000 games
3 min/game * 3000 games => over 150 hours (>6 days 24/7)
This kind of "little" test should be planned for later...

[rulleeeee]

You're right

And 100 games per opening?

300 games in total

[deeds]

300 or 3000 games is not important.
No, the main problem is that my servers are already busy for multi-day tasks.
Either I do the test in good conditions, Either I do it later.
In addition, we are not in a hurry and we have nothing to gain from it.

server #1 - task @ 24% (of one week) :


server #2 - task @ 41% (of one month) :


server #3 - task @ 50% (of 9 days) :

[rulleeeee]

No problem. just a test out of curiosity