Frithjof

What about https://github.com/apertureless/vue-cookie-law?

Hi @jesellier, Nice proposal, thanks for writing this up. This seems like a natural evolution of skpro. I like the idea to make the loss functions callable objects. Once the...

Thanks! 1. That's interesting, either way it will be fine 2. Makes sense. 3. I think that's a good way to put, although, there is some cost to it, e.g....

I think having `location`/`scale` as generic and semantically vague default and optional derived `mean`/`variance` that may or may not exist is a good idea.

I think removing would weaken the purpose of the distribution interface. After all, you want to type `my_distribution.` and have your IDE autocomplete `my_distribution.mean()` for you. How about we just...

> How would that auto-complete even work, @frthjf? Based on what would one compute the mean? If the method is defined by the base class, it would be autocompleted via...

Ok, makes sense! I'm still not fully convinced of the Mixin approach. I prefer a ducktyping approach over explicit declaration, but I think that's just a matter of taste, so...

I've been working with TFP for a while and I agree it has a good feature set (after all, PyMC switched to it as a backend as well). One big...

Looking a bit more closely, if I change the trigger function to ```python def trigger(self, event: str, *args: T.Any, **kw: T.Any) -> list: callbacks = list(self._events.get(event, [])) return [callback(*args, **kw)...

Hi @honestee, sorry to hear that you've encountered issues. As a workaround you should be able to do something like this: ``` import six import sys sys.modules['sklearn.externals.six'] = six import...