DEPR: scalar index for length-1-list level groupby #51817
Conversation
Well that's very reasonable, should have thought of that... What's the best way to go about this? Maybe just make a new PR (with an actual "depreciation") and then leave/close this branch so that it can be used for reference when the time is right? |
|
I would suggest just modifying this PR to deprecate the behavior; as you've already demonstrated the changes to enforce that deprecation won't be extensive. |
…august-tengland/pandas-august-tengland into single-element-level-argument-iteration
|
I'm not sure how to resolve this failing mypy check, using |
|
Looks okay to add a type ignore here. We should eventually use a type guard with is_list_like but might now be able to do that just yet. |
| @pytest.mark.parametrize( | ||
| "level_arg, multiindex", [([0], False), ((0,), False), ([0], True), ((0,), True)] | ||
| ) |
There was a problem hiding this comment.
This is a small enough case that it is perfectly fine as-is, but wanted to mention that you can stack parameterizations and it will take the cartesian product:
@pytest.mark.parametrize("level_arg", [[0], (0,)])
@pytest.mark.parametrize("multiindex", [False, True])
|
Thanks @august-tengland! |
groupbywith multi-index #51583doc/source/whatsnew/vX.X.X.rstfile if fixing a bug or adding a new feature.Previously, iterating over a Groupby object that had been initialized with a length-1-list
levelparameter (ex. level = [0]) would return scalar indexes as opposed to tuples with only the first element defined (ex. 0 as opposed to (0, ) ). This behavior differs from when using thebyparameter and so the change is motivated.