BUG/API Raise ValueError for non-unique stack by name #6738
Conversation
|
|
||
|
|
||
| def _reference_duplicate_name(names, name): | ||
| """ |
There was a problem hiding this comment.
I would make this a method of Multiindex
There was a problem hiding this comment.
That's what I was thinking.
I also changed the implementation a bit. My way was buggy since it would raise whenever there was duplicates and a name was in names (even if name wasn't duplicated). But I want it to only raise when name is one of the duplicates.
|
@TomAugspurger my expectation would be that index names should be hashable. They're supposed to be key-esque, right? (plus tuples are still hashable). If you needed to, I'd catch the TypeError from trying to hash non-hashable object and just warn in that case. |
|
@TomAugspurger pls rebase....seems ok otherwise |
|
@TomAugspurger this closes #6319 as well? |
Should raise a ValueError when (un)stacking a DataFrame on a nonunique level. Previous behavior was to raise a KeyError (not deliberately). Closes pandas-dev#6729.
|
This only closes #6319 if we don't want to support pivots/crosstabs where the names are the same. But we should be able to support them. I can do a proper fix for #6319 for .14 probably. I haven't merged this because my pushes to the remote branch aren't showing up here. All I need to do is rebase, so I may just do that as I'm merging. Should just be release notes / v0.14 that have conflicts. Sound OK? |
|
how so you push to the remote? |
|
Somewhere along the way I created two branches: Making a new PR at #6849 |
Closes #6729
We should raise a ValueError when (un)stacking and referring to a non-unique index name. You can still stack and unstack by position like normal (no ambiguity there, unless your names are integers and that integer is duplicated, in which case we'll raise a ValueError).
Right now I'm assuming that index names are always hashable, but that's not necessarily true, is it? Any good ways to check the uniqueness of the names? I didn't see anything on MultiIndex that did this.