MNT clean futurewarning for 1.0 | _deprecate_positional_args

[TomDLT]

Remove @_deprecate_positional_args for 1.0 (part of #19335).
These decorators have been added in 0.23 (see #15005).

Now that positional arguments are not allowed, we can safely reorder parameters when appropriate (maybe better in a separate issue though).

Commands used

git ls-files | xargs sed -i -e ':a;N;$!ba;s/        @_deprecate_positional_args\n//g'
git ls-files | xargs sed -i -e ':a;N;$!ba;s/    @_deprecate_positional_args\n//g'
git ls-files | xargs sed -i -e ':a;N;$!ba;s/@_deprecate_positional_args\n//g'
git ls-files | xargs sed -i -e ':a;N;$!ba;s/from ...utils.validation import _deprecate_positional_args\n//g'
git ls-files | xargs sed -i -e ':a;N;$!ba;s/from ..utils.validation import _deprecate_positional_args\n//g' 
git ls-files | xargs sed -i -e ':a;N;$!ba;s/from .utils.validation import _deprecate_positional_args\n//g' 
git ls-files | xargs sed -i -e ':a;N;$!ba;s/from .validation import _deprecate_positional_args\n//g' 
git ls-files | xargs sed -i -e ':a;N;$!ba;s/, _deprecate_positional_args//g'  
git ls-files | xargs sed -i -e ':a;N;$!ba;s/import _deprecate_positional_args,/import/g' 
# Then manual checks with 
git grep "_deprecate_positional_args"

[TomDLT]

I simply silenced mypy errors by adding # type: ignore here and there.

mypy errors

sklearn/decomposition/tests/test_sparse_pca.py:192: error: List item 0 has incompatible type "Type[SparsePCA]"; expected "Type[SparsePCA]"
sklearn/feature_selection/_univariate_selection.py:853: error: Dict entry 0 has incompatible type "str": "Type[SelectPercentile]"; expected "str": "Type[_BaseFilter]"
sklearn/feature_selection/_univariate_selection.py:854: error: Dict entry 1 has incompatible type "str": "Type[SelectKBest]"; expected "str": "Type[_BaseFilter]"
sklearn/svm/tests/test_svm.py:530: error: List item 1 has incompatible type "Tuple[Type[NuSVC], str]"; expected "Tuple[Type[BaseSVC], str]"
sklearn/svm/tests/test_svm.py:550: error: List item 1 has incompatible type "Tuple[Type[NuSVC], Dict[str, List[float]]]"; expected "Tuple[Type[BaseSVC], Dict[str, List[float]]]"
sklearn/svm/tests/test_svm.py:1133: error: List item 0 has incompatible type "Type[SVC]"; expected "Type[BaseSVC]"
sklearn/svm/tests/test_svm.py:1144: error: List item 0 has incompatible type "Type[SVC]"; expected "Type[BaseSVC]"
sklearn/model_selection/tests/test_split.py:347: error: List item 0 has incompatible type "Type[StratifiedKFold]"; expected "Type[_BaseKFold]"
sklearn/model_selection/tests/test_split.py:372: error: List item 0 has incompatible type "Type[StratifiedKFold]"; expected "Type[_BaseKFold]"
sklearn/model_selection/tests/test_split.py:407: error: List item 0 has incompatible type "Type[StratifiedKFold]"; expected "Type[_BaseKFold]"
sklearn/model_selection/tests/test_split.py:450: error: List item 0 has incompatible type "Type[KFold]"; expected "Type[_BaseKFold]"
sklearn/model_selection/tests/test_split.py:450: error: List item 1 has incompatible type "Type[StratifiedKFold]"; expected "Type[_BaseKFold]"
sklearn/tree/tests/test_tree.py:68: error: Argument 1 to "update" of "dict" has incompatible type "Dict[str, Type[DecisionTreeRegressor]]"; expected "Mapping[str, Type[DecisionTreeClassifier]]"
sklearn/decomposition/tests/test_dict_learning.py:606: error: List item 0 has incompatible type "Type[DictionaryLearning]"; expected "Type[_BaseSparseCoding]"
sklearn/ensemble/tests/test_gradient_boosting.py:1225: error: List item 2 has incompatible type <tuple: 3 items>; expected "Tuple[ABCMeta, function, Type[MultiOutputMixin]]"
Found 15 errors in 7 files (checked 499 source files)

[amueller]

You're not removing the actual helper, right? Wouldn't it be good to do that, so we can make sure we got them all?

[TomDLT]

You're not removing the actual helper, right?

The helper is still there, it is required until 1.1 for this function:

scikit-learn/sklearn/inspection/_plot/partial_dependence.py

Lines 823 to 825 in a9cc0ed

	@_deprecate_positional_args(version="1.1")
	def plot(self, *, ax=None, n_cols=3, line_kw=None, contour_kw=None):
	"""Plot partial dependence plots.

so we can make sure we got them all?

Result of git grep @_deprecate_positional_args on this branch:

sklearn/inspection/_plot/partial_dependence.py:    @_deprecate_positional_args(version="1.1")
sklearn/utils/tests/test_validation.py:    @_deprecate_positional_args
sklearn/utils/tests/test_validation.py:    @_deprecate_positional_args
sklearn/utils/tests/test_validation.py:    @_deprecate_positional_args
sklearn/utils/tests/test_validation.py:    @_deprecate_positional_args(version="1.1")
sklearn/utils/tests/test_validation.py:        @_deprecate_positional_args
sklearn/utils/tests/test_validation.py:        @_deprecate_positional_args

[ogrisel]

I pushed a commit to bump the default version value to "1.1 (renaming of 0.26)" in the private _deprecate_positional_args helper to avoid confusion. I also checked and the only need is the for the PDP function cited in the parent issue.

@thomasjpfan the mypy linter now complains with:

sklearn/neighbors/tests/test_neighbors.py:95: error: List item 0 has incompatible type "Type[KNeighborsClassifier]"; expected "Type[NeighborsBase]"
sklearn/neighbors/tests/test_neighbors.py:96: error: List item 1 has incompatible type "Type[KNeighborsRegressor]"; expected "Type[NeighborsBase]"
sklearn/neighbors/tests/test_neighbors.py:1749: error: List item 0 has incompatible type "Type[KNeighborsClassifier]"; expected "Type[NeighborsBase]"
sklearn/neighbors/tests/test_neighbors.py:1750: error: List item 1 has incompatible type "Type[KNeighborsRegressor]"; expected "Type[NeighborsBase]"

I could probably just silence it but do you understand what is causing this? I checked and KNeighborsClassifier inherits from NeighborsBase.

[thomasjpfan]

My sense is that decorators + mypy can give strange results. In this case, adding pytest to the azure CI linting environment resolved the issue. (pytest 6.x added better typing support)

[ogrisel]

+1 for merging this PR then. Thanks @thomasjpfan for the mypy investigation / fix.

[ogrisel]

This will make our tracebacks and profiler report simpler and easier to understand.

[thomasjpfan]

LGTM