Why is SF6 so much stronger?

[clumma]

How did they make such a leap in one version? The code is available. Is the cause not understood? I looked here and on other forums, and didn't see discussion on this question.

-Carl

[Dann Corbit]

How did they make such a leap in one version? The code is available. Is the cause not understood? I looked here and on other forums, and didn't see discussion on this question.

-Carl

There were a zillion min-versions between the two major versions.
Look here:
http://abrok.eu/stockfish/

[lucasart]

How did they make such a leap in one version? The code is available. Is the cause not understood? I looked here and on other forums, and didn't see discussion on this question.

-Carl

maybe this will give you a hint:
http://tests.stockfishchess.org/tests
http://tests.stockfishchess.org/users
https://groups.google.com/forum/?fromgr ... ishcooking

[Uri Blass]

How did they make such a leap in one version? The code is available. Is the cause not understood? I looked here and on other forums, and didn't see discussion on this question.

-Carl

I think that people do not understand the reason.
The target of testing in the stockfish framework is simply to make stockfish stronger and not understand why it is so strong(otherwise they could remove the non important parts that only give 1 or 2 elo) and make comments in the source about every code that remains how much elo it gives based on testing.

I do not believe that it is so strong because of too many additions that give only 1-2 elo and my guess is that if we remove everything that does not give more than 2 elo we are going to remain with something that is less than 200 elo weaker and less than half of the size of stockfish.

Of course you need a lot of testing to do it and people prefer to do testing for improving stockfish and not to evaluate the value of every code that is in it.

Edit:I thought about the question why stockfish is so strong and not why stockfish6 is stronger than stockfish5 but the same is for stockfish6 relative to stockfish5 and there is no testing of 200,000 games to find the elo value of every patch that they accept to get a good estimate of the value of the patch so we practically do not know the patches that are more important and the patches that are less important.

[clumma]

Thanks for the link Dann, it gives slightly more information than github. Still nothing resembling an analysis of the improvement though.

-Carl

[clumma]

Yes, I know about the testing protocol. You also pointed me at the forum. Is this supposed to make me believe you know the answer to my question?

There happens to be a thread, "+50 ELO, MY GOD", where a 50 Elo improvement is apparently attributed to a single change to the SMP logic. But SF6 is much stronger even on 1 core.

It seems SF team are making progress but don't understand how.

[Joerg Oster]

Yes, I know about the testing protocol. You also pointed me at the forum. Is this supposed to make me believe you know the answer to my question?

There happens to be a thread, "+50 ELO, MY GOD", where a 50 Elo improvement is apparently attributed to a single change to the SMP logic. But SF6 is much stronger even on 1 core.

It seems SF team are making progress but don't understand how.

On github you can follow every single commit.
So what do you not understand?

[BBauer]

It seems SF team are making progress but don't understand how.

Now need to have a deeper metaphysical understanding.

The apple falls down from the tree. In some sense we know why even so gravity is not fully understood.

Kind regards
Bernhard

[cdani]

It seems SF team are making progress but don't understand how.

The best answer I can think is to compare SF to a Grand Master who is a lot better thanks to the addition of a lot of concepts. Some concepts gives more elo to him, and some less.

[clumma]

What is the distribution of Elo improvement over commits since version 5? All equal and small? If not, what did the main changes accomplish (SMP handling, fine tuning of parameters...)?

If answer is fine tuning of parameters I would be surprised, since I expect this to yield diminishing returns by now.

What happens if commits are applied in a different order?

Etc.

Why is it not obvious what "understanding" means?

-Carl