Lczero stalled in progress?

[syzygy]

Yes, Milos and his negativity is now ignored to me. But shouldn't he be banned now ?

I don't know what he wrote this time, but in general my answer to your question is yes. His contribution to the forum is strictly negative.

[Leo]

You don't have the authority to ban him. BTY how do I ignore someone in here?

[syzygy]

You don't have the authority to ban him. BTY how do I ignore someone in here?

Right-click on the person to view his profile and click on "Add foe".

[mirek]

Because infinite growth is infeasible and saturation always happens sooner or later.
What you are actually asking is why progress is really not in line with expectations of so many ppl, but that is kind of question for psychologist so you kind of posed it in a wrong forum .

Seems you quite often post conclussions that directly contradict the Deepmind paper, so maybe that could be even more interesting puzzle for a psychologist.

So by your logic anyone that doesn't esteem that Google advertising preprint as bible is crazy? Interesting opinion.

Let me rephrase the question: so by your logic anyone who expects the Deepmind results can and will be replicated is crazy?

[Albert Silver]

Those are pretty natural terms, if you don't understand it I can't help you.

You do realize that his comment was in response to my report about the bug, and that your reply makes no sense whatsoever in the context. As such I'm not the one in need of help.

There is no more threaded view. I replied to what is in the comment, i.e. question why that is related to the progress stalled from thread title, I really don't care who replied to whom originally and for what.

Ok, now I understand. I had thought your replies were somehow related to what the post you were quoting. I had not realized they were just random posts.

[lantonov]

Obviously the move is 27. Qxd7 which Leela fails to play when given the history. If shown just the FEN, it finds it. Tests of previous neural network versions show the following probability to play Qxd7, as reported in the GitHub:

27. Qxd7 or 27. Rxd7 ?
Wish Leela reversing the downward trend ASAP. Going back to some earlier point is always an option.

[corres]

[quote=syzygy post_id=761798 time=1526002093 user_id=6292]
The problems started on 3 May, which was when v0.8 was released. So it seems rather likely that a severe bug was introduced in one of the about 100 commits between v0.7 and v0.8.
[/quote]

I also think the issues were introduced by v0.8 and the further versions.
But these were not some serious bugs but a conceptional fault.
They made a lot of changes on the code of lczero simultaneously without any evidence about that every modifications have positive effect to the Elo of LC0.
In my opinion they should return to bug fixed v0.7 and after the Elo stabilized (may be enhanced) they should re-introduced the modifications one by one. In this mode the bad changing would be cleared up.

[Werewolf]

The problems started on 3 May, which was when v0.8 was released. So it seems rather likely that a severe bug was introduced in one of the about 100 commits between v0.7 and v0.8.

I also think the issues were introduced by v0.8 and the further versions.
But these were not some serious bugs but a conceptional fault.
They made a lot of changes on the code of lczero simultaneously without any evidence about that every modifications have positive effect to the Elo of LC0.
In my opinion they should return to bug fixed v0.7 and after the Elo stabilized (may be enhanced) they should re-introduced the modifications one by one. In this mode the bad changing would be cleared up.

This seems logical to me.

[Robert Flesher]

I am just curious if the experts can explain why Lczero seems to have stalled in progress?

A massive bug was revealed today, which has also affected the training. Leela will literally refuse to capture a hanging queen:



https://groups.google.com/forum/#!topic ... iOlKkuhCGw

It seems the stalled progress has still not been fixed and the program seems to be losing ELO. I am still very interested, thus hoping the issue is found.

[Robert Flesher]

The problems started on 3 May, which was when v0.8 was released. So it seems rather likely that a severe bug was introduced in one of the about 100 commits between v0.7 and v0.8.

I also think the issues were introduced by v0.8 and the further versions.
But these were not some serious bugs but a conceptional fault.
They made a lot of changes on the code of lczero simultaneously without any evidence about that every modifications have positive effect to the Elo of LC0.
In my opinion they should return to bug fixed v0.7 and after the Elo stabilized (may be enhanced) they should re-introduced the modifications one by one. In this mode the bad changing would be cleared up.

This seems logical to me.

'
+1