Maximilian Linhoff
Maximilian Linhoff
NaN is not usable for missing data. This was a major design flaw in pandas that was rectified with version 1.0. Please don't repeat mistakes that others have already made...
Quantities can have integer dtype.
> However, if we decide that get_frame_attr_names() serves no purpose anymore then it should be deprecated properly. I would prefer the latter. What would need to be done for this?
Ok, so you mean I should leave the old method untouched, replace internal usage with just using `frame_attributes` and add the `get_frame_attr_defaults` method?
Sorry, but looking at the places this is used and at the comments, it seems like this is a major api change.... not just the simple performance improvement I had...
Speedups with respect to master are still about the same: ``` ❯ ipython -i time_attrs.py Python 3.9.12 | packaged by conda-forge | (main, Mar 24 2022, 23:22:55) Type 'copyright', 'credits'...
@eerovaher I think I addressed all comments and squashed into a single commit.
The codecov change is rather disturbing... (-10%), is that unrelated???
Ah nevermind, it seems codecov is uploaded from a partial test first and then when the full test suite finishes it's overwritten...
@eerovaher should this have been merged automatically? I saw you added the Label to merge after ci passes