Yuri Baburov

@ramreddymettu I don't remember. Things I think I could have tried: 1) move to Julia 0.4 2) recompiling 3) identifying and manually fixing wrong places in the source code. One...

Yes, I'm trying to do L2 penalization. By "input" i mean hidden / non-final layers. By values I mean weights themselves. Something like http://lasagne.readthedocs.org/en/latest/modules/regularization.html Ideally if I could sum up...

I believe this is caused by an internal trie creation on model loading, which then stays in memory and consumes a lot of RAM. Mozilla in their version saves this...

Эм.. А как в Зализняке от odict найти ударение для конкретной словоформы? Хорошо было бы просто сделать отдельный словарь с ударениями. У меня такой даже был где-то, кажется, на 2М...

@EndingCredits Using raw names still isn't good. You seemed to ignore all important points in my message. Someone will write t['neme1'] and it won't work, because it's t['name1']. t[t.name1] is...

@EndingCredits it is much better when linter/compiler warns you about this, instead of at run-time.

@srush for that you might have x.batch[1] as well, and anyway it's no worse than x[{"batch": 1}] .

Please add a column "reproducible" then, I think everyone will agree this is the right step to scale this up.

1) you have only one python version installed so for example you shouldn't ever consider "python_version >= 3.10" alternative when you have python 3.8 installed. Next, after you checked for...

@finswimmer thanks. That's ok, you can still simplify it (Python >=3.8 & Python = 1.17.3 & Python>=3.7) | (Numpy >=1.19.3 & Python >= 3.9) | ...) ==> (numpy >= 1.19.3)...