BUG: groupby.rank for dt64tz, period dtypes #38187
Conversation
| @@ -491,8 +491,11 @@ def _ea_wrap_cython_operation( | |||
| res_values, names = self._cython_operation( | |||
| kind, values, how, axis, min_count, **kwargs | |||
| ) | |||
| if how in ["rank"]: | |||
There was a problem hiding this comment.
hmm, this is prob not the best location for this longer term (eg.. we should have mappings from ops -> whether they should coerce back or not).
There was a problem hiding this comment.
yah, we have something like that in the calling class. im in the process of moving all of this wrapping to the lowest level possible (which luckily coincides with it being in the fewest places possible), xref #37494 (comment)
There was a problem hiding this comment.
Could this also be handled in maybe_cast_result_dtype? (which basically is somewhat implementing a mapping from ops -> expected dtype)
There was a problem hiding this comment.
potentially, yah. this calls simple_new instead of from_sequence, which makes it a bit of an outlier.
there's also groupby.base.cython_cast_blocklist that i think is intended for something along these lines
|
hmm, looks like CI is hanging, can you merge master and pin gon green. |
not merged master, but re-run an green |
black pandasgit diff upstream/master -u -- "*.py" | flake8 --diffThis sits on top of #38162