REF: implement putmask for CI/DTI/TDI/PI #36400
Conversation
jreback
added
Refactor
| try: | ||
| code_value = self._data._validate_where_value(value) | ||
| except (TypeError, ValueError): | ||
| return self.astype(object).putmask(mask, value) |
| codes = self._data._ndarray.copy() | ||
| np.putmask(codes, mask, code_value) | ||
| cat = self._data._from_backing_data(codes) | ||
| return type(self)._simple_new(cat, name=self.name) |
There was a problem hiding this comment.
looks like we have coverage for all of the datetimelike but none of the categorical; will update
| @@ -422,6 +422,17 @@ def where(self, cond, other=None): | |||
| cat = Categorical(values, dtype=self.dtype) | |||
| return type(self)._simple_new(cat, name=self.name) | |||
|
|
|||
| def putmask(self, mask, value): | |||
There was a problem hiding this comment.
not necessarily for today. but is there any value to pushing this down to the array and having a putmask_compat until NEP18 can be supported?
putmask on the Index returns a copy whereas putmask compat on the array would be expected to be inplace. This may not be so easy for Categorical, but for other numpy backed arrays could be more trivial.
also is the goal of extension array backed indexes to allow 3rd party EAs in the Index. If so, putmask on the array would need to be added to the EA interface?
There was a problem hiding this comment.
[...] putmask on the array would need to be added to the EA interface?
I would be in favor of this
There was a problem hiding this comment.
yep I am also +1 on this as this is a 'standard' array method, can you create an issue (we might have one?)
|
merging but if possible to followon for a) additional testing and b) any perf issues and c) EA api |
Avoids casting to ndarray which in some cases means an object-dtype cast.