Randomization fix

[gumityolcu]

Hi Dilya,

I made small fixes in the model randomization metric :

1- The state_dict for the metric now includes state_dict of the randomized model
2- I opted not to use make_func because I don't exactly understand why but functools.partial doesn't work for me. I think you have to fix parameters starting from the final parameter, which means it's not very useful imo. Specifically the code below does not work:

from functools import partial


def orig(one, two, three, four):
    print(one)
    print(two)
    print(three)
    print(four)


part = partial(orig, one=1, three=3)

part(23, 23)

[dilyabareeva]

Hi @gumityolcu, it doesn't work because you are supposed to use keyword arguments like this:

part(two=23, four=23)

[dilyabareeva]

on that topic, it would be a good practice to use keyword arguments everywhere, because it makes the code more readable and easier to debug. I'm not super consistent on this either, but I think we should get into the habit!

[gumityolcu]

Hi @gumityolcu, it doesn't work because you are supposed to use keyword arguments like this:

part(two=23, four=23)

I actually prefer that and that's the first thing I wrote initially. Now I figured out that we had a miniscule bug in make_func 👍🏼

So now i fixed that (i added 2 characters but you should still check to quickly make sure)
I integrated make_func into the metric. I added a bunch of stuff to the state_dict to be able to save and load the exact same object.

Finally, I noticed that I never gave the seed to the generator object so i fixed that (and the relevant functionality in saving and loading state_dicts)

[dilyabareeva]

I'd say ready to merge!