ENH: Add support for numpy 2's string dtype #58394
Conversation
| ndarray[uint8_int64_object_t] out, | ||
| ndarray[object] arr, | ||
| ndarray out, | ||
| ndarray arr, |
There was a problem hiding this comment.
This will make iteration slower.
Look into numpy C API iteration functions.
(e.g. look at Validator._validate)
If this gives perf improvement, maybe can be split out.
|
|
||
| @classmethod | ||
| def _from_sequence(cls, scalars, *, dtype: Dtype | None = None, copy: bool = False): | ||
| na_mask, any_na = libmissing.isnaobj(np.array(scalars, dtype=object), check_for_any_na=True) |
There was a problem hiding this comment.
Maybe can use _lib.convert_nans_to_NA instead?
Would get rid of diff in missing.pyx.
| return super()._str_endswith(pat, na) | ||
| pat = np.asarray(pat, dtype=get_numpy_string_dtype_instance()) | ||
| result = np.strings.endswith(self._ndarray, pat) | ||
| return BooleanArray(result, isna(self._ndarray)) |
|
|
||
| if ret.dtype == object and ret_dtype is not None: | ||
| # cast from object back to StringDType | ||
| return ret.astype(ret_dtype, copy=False) |
There was a problem hiding this comment.
In map_infer_mask there is a convert=True argument, that calls ``maybe_convert_objects on it.
Alternately, we could look at setting the dtype argument to an instance of StringDType, and hope that PyArray_SETITEM will work on it?
|
|
||
| def get_numpy_string_dtype_instance( | ||
| na_object=libmissing.NA, |
There was a problem hiding this comment.
Is na_object parameter used?
| ) | ||
| raise TypeError(msg) | ||
| dtype = self._inferred_dtype | ||
| if dtype not in allowed_types: |
There was a problem hiding this comment.
Are numpy dtypes hashable?
Might just want to add to allowed_types
| @@ -1479,7 +1481,7 @@ def find_common_type(types): | |||
| if t.kind in "iufc": | |||
| return np.dtype("object") | |||
|
|
|||
| return np_find_common_type(*types) | |||
| return np.result_type(*types) | |||
There was a problem hiding this comment.
Note to self: This (other changes in this file) are causing test failures.
| if self.values.dtype.kind == 'T': | ||
| return NumpyStringArray(self.values) | ||
| else: | ||
| return NumpyExtensionArray(self.values) |
There was a problem hiding this comment.
This is because NumpyExtensionArray inherits from ObjectStringArrayMixin, so unless I make this change there's no way to call the string method overrides in NumpyStringArray.
doc/source/whatsnew/vX.X.X.rstfile if fixing a bug or adding a new feature.Just submitting this for CI purposes, nothing to review yet.