CLN: Consistent and Annotated Return Type of _iterate_slices #28958
Conversation
pandas/core/groupby/generic.py
Outdated
| slice_axis = self._selection_list | ||
| slicer = lambda x: self.obj[x] | ||
| def _iterate_slices(self) -> Iterator[Tuple[Hashable, Series]]: | ||
| obj = self._selected_obj.copy() |
There was a problem hiding this comment.
Mistake on my end. Didn’t want to mutate passed frame in the axis=1 case but I can get rid of this
There was a problem hiding this comment.
Off topic: can we make the axis=1 case unnecessary by just transposing when we see it? Not sure exactly how that would work, but it would let us get rid of some lambda-wrapping, which would be nice.
|
Might want to look at DataFrameGroupBy._transform_item_by_item. If I'm reading it correctly, that will screw up if we have non-unique columns. (I think that's relevant to what you're doing here) |
Ran entire GroupBy benchmark and showed no changes
Yea looks it could be consolidated to use this, though leaving to a follow up Also w.r.t. your comment on axis=1 case seems logical but might be some complications around managing selection state with that and how things are constructed. May take a look down the road, though if you see a way to make it work in the meantime makes sense to me |
|
thanks |
General pre-cursor to getting block management out of groupby. This is also a pre-cursor to fixing #21668 but needs to be coupled with a few more changes as a follow up
On master calls to _iterate_slices look up by label, potentially yielding a DataFrame if there were duplicated columns. This takes the surprise out of that and simply returns a Tuple of label / series for each item along the axis
@jbrockmendel