From fed9115e18c17d76e1aec35ed0fbc99aec272611 Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 21:55:04 +0100 Subject: [PATCH 01/10] MAINT use sklearn_compat for multi-version scikit-learn compatibilities --- python-package/lightgbm/compat.py | 98 +--- python-package/lightgbm/sklearn_compat.py | 643 ++++++++++++++++++++++ 2 files changed, 649 insertions(+), 92 deletions(-) create mode 100644 python-package/lightgbm/sklearn_compat.py diff --git a/python-package/lightgbm/compat.py b/python-package/lightgbm/compat.py index 96dee6522572..9a3f3a248b6a 100644 --- a/python-package/lightgbm/compat.py +++ b/python-package/lightgbm/compat.py @@ -12,91 +12,11 @@ from sklearn.preprocessing import LabelEncoder from sklearn.utils.class_weight import compute_sample_weight from sklearn.utils.multiclass import check_classification_targets - from sklearn.utils.validation import assert_all_finite, check_array, check_X_y - - try: - from sklearn.exceptions import NotFittedError - from sklearn.model_selection import BaseCrossValidator, GroupKFold, StratifiedKFold - except ImportError: - from sklearn.cross_validation import BaseCrossValidator, GroupKFold, StratifiedKFold - from sklearn.utils.validation import NotFittedError - try: - from sklearn.utils.validation import _check_sample_weight - except ImportError: - from sklearn.utils.validation import check_consistent_length - - # dummy function to support older version of scikit-learn - def _check_sample_weight(sample_weight: Any, X: Any, dtype: Any = None) -> Any: - check_consistent_length(sample_weight, X) - return sample_weight - - try: - from sklearn.utils.validation import validate_data - except ImportError: - # validate_data() was added in scikit-learn 1.6, this function roughly imitates it for older versions. - # It can be removed when lightgbm's minimum scikit-learn version is at least 1.6. - def validate_data( - _estimator, - X, - y="no_validation", - accept_sparse: bool = True, - # 'force_all_finite' was renamed to 'ensure_all_finite' in scikit-learn 1.6 - ensure_all_finite: bool = False, - ensure_min_samples: int = 1, - # trap other keyword arguments that only work on scikit-learn >=1.6, like 'reset' - **ignored_kwargs, - ): - # it's safe to import _num_features unconditionally because: - # - # * it was first added in scikit-learn 0.24.2 - # * lightgbm cannot be used with scikit-learn versions older than that - # * this validate_data() re-implementation will not be called in scikit-learn>=1.6 - # - from sklearn.utils.validation import _num_features - - # _num_features() raises a TypeError on 1-dimensional input. That's a problem - # because scikit-learn's 'check_fit1d' estimator check sets that expectation that - # estimators must raise a ValueError when a 1-dimensional input is passed to fit(). - # - # So here, lightgbm avoids calling _num_features() on 1-dimensional inputs. - if hasattr(X, "shape") and len(X.shape) == 1: - n_features_in_ = 1 - else: - n_features_in_ = _num_features(X) - - no_val_y = isinstance(y, str) and y == "no_validation" - - # NOTE: check_X_y() calls check_array() internally, so only need to call one or the other of them here - if no_val_y: - X = check_array( - X, - accept_sparse=accept_sparse, - force_all_finite=ensure_all_finite, - ensure_min_samples=ensure_min_samples, - ) - else: - X, y = check_X_y( - X, - y, - accept_sparse=accept_sparse, - force_all_finite=ensure_all_finite, - ensure_min_samples=ensure_min_samples, - ) - - # this only needs to be updated at fit() time - _estimator.n_features_in_ = n_features_in_ - - # raise the same error that scikit-learn's `validate_data()` does on scikit-learn>=1.6 - if _estimator.__sklearn_is_fitted__() and _estimator._n_features != n_features_in_: - raise ValueError( - f"X has {n_features_in_} features, but {_estimator.__class__.__name__} " - f"is expecting {_estimator._n_features} features as input." - ) - - if no_val_y: - return X - else: - return X, y + from sklearn.utils.validation import assert_all_finite + from sklearn.exceptions import NotFittedError + from sklearn.model_selection import BaseCrossValidator, GroupKFold, StratifiedKFold + from sklearn.utils.validation import _check_sample_weight + from .sklearn_compat.utils.validation import validate_data SKLEARN_INSTALLED = True _LGBMBaseCrossValidator = BaseCrossValidator @@ -144,13 +64,7 @@ class _LGBMRegressorBase: # type: ignore # additional scikit-learn imports only for type hints if TYPE_CHECKING: - # sklearn.utils.Tags can be imported unconditionally once - # lightgbm's minimum scikit-learn version is 1.6 or higher - try: - from sklearn.utils import Tags as _sklearn_Tags - except ImportError: - _sklearn_Tags = None - + from .sklearn_compat.utils import Tags as _sklearn_Tags """pandas""" try: diff --git a/python-package/lightgbm/sklearn_compat.py b/python-package/lightgbm/sklearn_compat.py new file mode 100644 index 000000000000..9ce6792203ce --- /dev/null +++ b/python-package/lightgbm/sklearn_compat.py @@ -0,0 +1,643 @@ +"""Ease developer experience to support multiple versions of scikit-learn. + +This file is intended to be vendored in your project if you do not want to depend on +`sklearn-compat` as a package. Then, you can import directly from this file. + +Be aware that depending on `sklearn-compat` does not add any additional dependencies: +we are only depending on `scikit-learn`. + +Version: 0.1.0 +""" + +from __future__ import annotations + +import platform +import sys +from dataclasses import dataclass, field +from typing import Callable, Literal + +import sklearn +from sklearn.utils._param_validation import validate_parameter_constraints +from sklearn.utils.fixes import parse_version + +sklearn_version = parse_version(parse_version(sklearn.__version__).base_version) + + +######################################################################################## +# The following code does not depend on the sklearn version +######################################################################################## + + +# parameters validation +class ParamsValidationMixin: + """Mixin class to validate parameters.""" + + def _validate_params(self): + """Validate types and values of constructor parameters. + + The expected type and values must be defined in the `_parameter_constraints` + class attribute, which is a dictionary `param_name: list of constraints`. See + the docstring of `validate_parameter_constraints` for a description of the + accepted constraints. + """ + if hasattr(self, "_parameter_constraints"): + validate_parameter_constraints( + self._parameter_constraints, + self.get_params(deep=False), + caller_name=self.__class__.__name__, + ) + + +# tags infrastructure +def _dataclass_args(): + if sys.version_info < (3, 10): + return {} + return {"slots": True} + + +def get_tags(estimator): + """Get estimator tags in a consistent format across different sklearn versions. + + This function provides compatibility between sklearn versions before and after 1.6. + It returns either a Tags object (sklearn >= 1.6) or a converted Tags object from + the dictionary format (sklearn < 1.6) containing metadata about the estimator's + requirements and capabilities. + + Parameters + ---------- + estimator : estimator object + A scikit-learn estimator instance. + + Returns + ------- + tags : Tags + An object containing metadata about the estimator's requirements and + capabilities (e.g., input types, fitting requirements, classifier/regressor + specific tags). + """ + try: + from sklearn.utils._tags import get_tags + + return get_tags(estimator) + except ImportError: + from sklearn.utils._tags import _safe_tags + + return _to_new_tags(_safe_tags(estimator), estimator) + + +def _to_new_tags(old_tags, estimator=None): + """Utility function convert old tags (dictionary) to new tags (dataclass).""" + input_tags = InputTags( + one_d_array="1darray" in old_tags["X_types"], + two_d_array="2darray" in old_tags["X_types"], + three_d_array="3darray" in old_tags["X_types"], + sparse="sparse" in old_tags["X_types"], + categorical="categorical" in old_tags["X_types"], + string="string" in old_tags["X_types"], + dict="dict" in old_tags["X_types"], + positive_only=old_tags["requires_positive_X"], + allow_nan=old_tags["allow_nan"], + pairwise=old_tags["pairwise"], + ) + target_tags = TargetTags( + required=old_tags["requires_y"], + one_d_labels="1dlabels" in old_tags["X_types"], + two_d_labels="2dlabels" in old_tags["X_types"], + positive_only=old_tags["requires_positive_y"], + multi_output=old_tags["multioutput"] or old_tags["multioutput_only"], + single_output=not old_tags["multioutput_only"], + ) + if estimator is not None and ( + hasattr(estimator, "transform") or hasattr(estimator, "fit_transform") + ): + transformer_tags = TransformerTags( + preserves_dtype=old_tags["preserves_dtype"], + ) + else: + transformer_tags = None + estimator_type = getattr(estimator, "_estimator_type", None) + if estimator_type == "classifier": + classifier_tags = ClassifierTags( + poor_score=old_tags["poor_score"], + multi_class=not old_tags["binary_only"], + multi_label=old_tags["multilabel"], + ) + else: + classifier_tags = None + if estimator_type == "regressor": + regressor_tags = RegressorTags( + poor_score=old_tags["poor_score"], + multi_label=old_tags["multilabel"], + ) + else: + regressor_tags = None + return Tags( + estimator_type=estimator_type, + target_tags=target_tags, + transformer_tags=transformer_tags, + classifier_tags=classifier_tags, + regressor_tags=regressor_tags, + input_tags=input_tags, + array_api_support=old_tags["array_api_support"], + no_validation=old_tags["no_validation"], + non_deterministic=old_tags["non_deterministic"], + requires_fit=old_tags["requires_fit"], + _skip_test=old_tags["_skip_test"], + ) + + +######################################################################################## +# Upgrading for scikit-learn 1.4 +######################################################################################## + + +if sklearn_version < parse_version("1.4"): + + def _is_fitted(estimator, attributes=None, all_or_any=all): + """Determine if an estimator is fitted + + Parameters + ---------- + estimator : estimator instance + Estimator instance for which the check is performed. + + attributes : str, list or tuple of str, default=None + Attribute name(s) given as string or a list/tuple of strings + Eg.: ``["coef_", "estimator_", ...], "coef_"`` + + If `None`, `estimator` is considered fitted if there exist an + attribute that ends with a underscore and does not start with double + underscore. + + all_or_any : callable, {all, any}, default=all + Specify whether all or any of the given attributes must exist. + + Returns + ------- + fitted : bool + Whether the estimator is fitted. + """ + if attributes is not None: + if not isinstance(attributes, (list, tuple)): + attributes = [attributes] + return all_or_any([hasattr(estimator, attr) for attr in attributes]) + + if hasattr(estimator, "__sklearn_is_fitted__"): + return estimator.__sklearn_is_fitted__() + + fitted_attrs = [ + v for v in vars(estimator) if v.endswith("_") and not v.startswith("__") + ] + return len(fitted_attrs) > 0 + +else: + from sklearn.utils.validation import _is_fitted # noqa: F401 + + +######################################################################################## +# Upgrading for scikit-learn 1.5 +######################################################################################## + + +if sklearn_version < parse_version("1.5"): + # chunking + # extmath + # fixes + from sklearn.utils import ( + _IS_32BIT, # noqa: F401 + _approximate_mode, # noqa: F401 + _in_unstable_openblas_configuration, # noqa: F401 + gen_batches, # noqa: F401 + gen_even_slices, # noqa: F401 + get_chunk_n_rows, # noqa: F401 + safe_sqr, # noqa: F401 + ) + from sklearn.utils import _chunk_generator as chunk_generator # noqa: F401 + + _IS_WASM = platform.machine() in ["wasm32", "wasm64"] + # indexing + # mask + # missing + # optional dependencies + # user interface + # validation + from sklearn.utils import ( + _determine_key_type, # noqa: F401 + _get_column_indices, # noqa: F401 + _print_elapsed_time, # noqa: F401 + _safe_assign, # noqa: F401 + _safe_indexing, # noqa: F401 + _to_object_array, # noqa: F401 + axis0_safe_slice, # noqa: F401 + check_matplotlib_support, # noqa: F401 + check_pandas_support, # noqa: F401 + indices_to_mask, # noqa: F401 + is_scalar_nan, # noqa: F401 + resample, # noqa: F401 + safe_mask, # noqa: F401 + shuffle, # noqa: F401 + ) + from sklearn.utils import _is_pandas_na as is_pandas_na # noqa: F401 +else: + # chunking + from sklearn.utils._chunking import ( + chunk_generator, # noqa: F401 + gen_batches, # noqa: F401 + gen_even_slices, # noqa: F401 + get_chunk_n_rows, # noqa: F401 + ) + + # indexing + from sklearn.utils._indexing import ( + _determine_key_type, # noqa: F401 + _get_column_indices, # noqa: F401 + _safe_assign, # noqa: F401 + _safe_indexing, # noqa: F401 + resample, # noqa: F401 + shuffle, # noqa: F401 + ) + + # mask + from sklearn.utils._mask import ( + axis0_safe_slice, # noqa: F401 + indices_to_mask, # noqa: F401 + safe_mask, # noqa: F401 + ) + + # missing + from sklearn.utils._missing import ( + is_pandas_na, # noqa: F401 + is_scalar_nan, # noqa: F401 + ) + + # optional dependencies + from sklearn.utils._optional_dependencies import ( # noqa: F401 + check_matplotlib_support, + check_pandas_support, # noqa: F401 + ) + + # user interface + from sklearn.utils._user_interface import _print_elapsed_time # noqa: F401 + + # extmath + from sklearn.utils.extmath import ( + _approximate_mode, # noqa: F401 + safe_sqr, # noqa: F401 + ) + + # fixes + from sklearn.utils.fixes import ( + _IS_32BIT, # noqa: F401 + _IS_WASM, # noqa: F401 + _in_unstable_openblas_configuration, # noqa: F401 + ) + + # validation + from sklearn.utils.validation import _to_object_array # noqa: F401 + +######################################################################################## +# Upgrading for scikit-learn 1.6 +######################################################################################## + + +if sklearn_version < parse_version("1.6"): + # test_common + from sklearn.utils.estimator_checks import _construct_instance + + def type_of_target(y, input_name="", *, raise_unknown=False): + # fix for raise_unknown which is introduced in scikit-learn 1.6 + from sklearn.utils.multiclass import type_of_target + + def _raise_or_return(target_type): + """Depending on the value of raise_unknown, either raise an error or + return 'unknown'. + """ + if raise_unknown and target_type == "unknown": + input = input_name if input_name else "data" + raise ValueError(f"Unknown label type for {input}: {y!r}") + else: + return target_type + + target_type = type_of_target(y, input_name=input_name) + return _raise_or_return(target_type) + + def _construct_instances(Estimator): + yield _construct_instance(Estimator) + + # validation + def validate_data(_estimator, /, **kwargs): + if "ensure_all_finite" in kwargs: + force_all_finite = kwargs.pop("ensure_all_finite") + else: + force_all_finite = True + return _estimator._validate_data(**kwargs, force_all_finite=force_all_finite) + + def _check_n_features(estimator, X, *, reset): + return estimator._check_n_features(X, reset=reset) + + def _check_feature_names(estimator, X, *, reset): + return estimator._check_feature_names(X, reset=reset) + + # tags infrastructure + @dataclass(**_dataclass_args()) + class InputTags: + """Tags for the input data. + + Parameters + ---------- + one_d_array : bool, default=False + Whether the input can be a 1D array. + + two_d_array : bool, default=True + Whether the input can be a 2D array. Note that most common + tests currently run only if this flag is set to ``True``. + + three_d_array : bool, default=False + Whether the input can be a 3D array. + + sparse : bool, default=False + Whether the input can be a sparse matrix. + + categorical : bool, default=False + Whether the input can be categorical. + + string : bool, default=False + Whether the input can be an array-like of strings. + + dict : bool, default=False + Whether the input can be a dictionary. + + positive_only : bool, default=False + Whether the estimator requires positive X. + + allow_nan : bool, default=False + Whether the estimator supports data with missing values encoded as `np.nan`. + + pairwise : bool, default=False + This boolean attribute indicates whether the data (`X`), + :term:`fit` and similar methods consists of pairwise measures + over samples rather than a feature representation for each + sample. It is usually `True` where an estimator has a + `metric` or `affinity` or `kernel` parameter with value + 'precomputed'. Its primary purpose is to support a + :term:`meta-estimator` or a cross validation procedure that + extracts a sub-sample of data intended for a pairwise + estimator, where the data needs to be indexed on both axes. + Specifically, this tag is used by + `sklearn.utils.metaestimators._safe_split` to slice rows and + columns. + """ + + one_d_array: bool = False + two_d_array: bool = True + three_d_array: bool = False + sparse: bool = False + categorical: bool = False + string: bool = False + dict: bool = False + positive_only: bool = False + allow_nan: bool = False + pairwise: bool = False + + @dataclass(**_dataclass_args()) + class TargetTags: + """Tags for the target data. + + Parameters + ---------- + required : bool + Whether the estimator requires y to be passed to `fit`, + `fit_predict` or `fit_transform` methods. The tag is ``True`` + for estimators inheriting from `~sklearn.base.RegressorMixin` + and `~sklearn.base.ClassifierMixin`. + + one_d_labels : bool, default=False + Whether the input is a 1D labels (y). + + two_d_labels : bool, default=False + Whether the input is a 2D labels (y). + + positive_only : bool, default=False + Whether the estimator requires a positive y (only applicable + for regression). + + multi_output : bool, default=False + Whether a regressor supports multi-target outputs or a classifier supports + multi-class multi-output. + + single_output : bool, default=True + Whether the target can be single-output. This can be ``False`` if the + estimator supports only multi-output cases. + """ + + required: bool + one_d_labels: bool = False + two_d_labels: bool = False + positive_only: bool = False + multi_output: bool = False + single_output: bool = True + + @dataclass(**_dataclass_args()) + class TransformerTags: + """Tags for the transformer. + + Parameters + ---------- + preserves_dtype : list[str], default=["float64"] + Applies only on transformers. It corresponds to the data types + which will be preserved such that `X_trans.dtype` is the same + as `X.dtype` after calling `transformer.transform(X)`. If this + list is empty, then the transformer is not expected to + preserve the data type. The first value in the list is + considered as the default data type, corresponding to the data + type of the output when the input data type is not going to be + preserved. + """ + + preserves_dtype: list[str] = field(default_factory=lambda: ["float64"]) + + @dataclass(**_dataclass_args()) + class ClassifierTags: + """Tags for the classifier. + + Parameters + ---------- + poor_score : bool, default=False + Whether the estimator fails to provide a "reasonable" test-set + score, which currently for classification is an accuracy of + 0.83 on ``make_blobs(n_samples=300, random_state=0)``. The + datasets and values are based on current estimators in scikit-learn + and might be replaced by something more systematic. + + multi_class : bool, default=True + Whether the classifier can handle multi-class + classification. Note that all classifiers support binary + classification. Therefore this flag indicates whether the + classifier is a binary-classifier-only or not. + + multi_label : bool, default=False + Whether the classifier supports multi-label output. + """ + + poor_score: bool = False + multi_class: bool = True + multi_label: bool = False + + @dataclass(**_dataclass_args()) + class RegressorTags: + """Tags for the regressor. + + Parameters + ---------- + poor_score : bool, default=False + Whether the estimator fails to provide a "reasonable" test-set + score, which currently for regression is an R2 of 0.5 on + ``make_regression(n_samples=200, n_features=10, + n_informative=1, bias=5.0, noise=20, random_state=42)``. The + dataset and values are based on current estimators in scikit-learn + and might be replaced by something more systematic. + + multi_label : bool, default=False + Whether the regressor supports multilabel output. + """ + + poor_score: bool = False + multi_label: bool = False + + @dataclass(**_dataclass_args()) + class Tags: + """Tags for the estimator. + + See :ref:`estimator_tags` for more information. + + Parameters + ---------- + estimator_type : str or None + The type of the estimator. Can be one of: + - "classifier" + - "regressor" + - "transformer" + - "clusterer" + - "outlier_detector" + - "density_estimator" + + target_tags : :class:`TargetTags` + The target(y) tags. + + transformer_tags : :class:`TransformerTags` or None + The transformer tags. + + classifier_tags : :class:`ClassifierTags` or None + The classifier tags. + + regressor_tags : :class:`RegressorTags` or None + The regressor tags. + + array_api_support : bool, default=False + Whether the estimator supports Array API compatible inputs. + + no_validation : bool, default=False + Whether the estimator skips input-validation. This is only meant for + stateless and dummy transformers! + + non_deterministic : bool, default=False + Whether the estimator is not deterministic given a fixed ``random_state``. + + requires_fit : bool, default=True + Whether the estimator requires to be fitted before calling one of + `transform`, `predict`, `predict_proba`, or `decision_function`. + + _skip_test : bool, default=False + Whether to skip common tests entirely. Don't use this unless + you have a *very good* reason. + + input_tags : :class:`InputTags` + The input data(X) tags. + """ + + estimator_type: str | None + target_tags: TargetTags + transformer_tags: TransformerTags | None = None + classifier_tags: ClassifierTags | None = None + regressor_tags: RegressorTags | None = None + array_api_support: bool = False + no_validation: bool = False + non_deterministic: bool = False + requires_fit: bool = True + _skip_test: bool = False + input_tags: InputTags = field(default_factory=InputTags) + + def _patched_more_tags(estimator, expected_failed_checks): + import copy + + from sklearn.utils._tags import _safe_tags + + original_tags = copy.deepcopy(_safe_tags(estimator)) + + def patched_more_tags(self): + original_tags.update({"_xfail_checks": expected_failed_checks}) + return original_tags + + estimator.__class__._more_tags = patched_more_tags + return estimator + + def check_estimator( + estimator=None, + generate_only=False, + *, + legacy: bool = True, + expected_failed_checks: dict[str, str] | None = None, + on_skip: Literal["warn"] | None = "warn", + on_fail: Literal["raise", "warn"] | None = "raise", + callback: Callable | None = None, + ): + # legacy, on_skip, on_fail, and callback are not supported and ignored + from sklearn.utils.estimator_checks import check_estimator + + return check_estimator( + _patched_more_tags(estimator, expected_failed_checks), + generate_only=generate_only, + ) + + def parametrize_with_checks( + estimators, + *, + legacy: bool = True, + expected_failed_checks: Callable | None = None, + ): + # legacy is not supported and ignored + from sklearn.utils.estimator_checks import parametrize_with_checks + + estimators = [ + _patched_more_tags(estimator, expected_failed_checks(estimator)) + for estimator in estimators + ] + + return parametrize_with_checks(estimators) + +else: + # test_common + # tags infrastructure + from sklearn.utils import ( + ClassifierTags, + InputTags, + RegressorTags, + Tags, + TargetTags, + TransformerTags, + ) + from sklearn.utils._test_common.instance_generator import ( + _construct_instances, # noqa: F401 + ) + from sklearn.utils.estimator_checks import ( + check_estimator, # noqa: F401 + parametrize_with_checks, # noqa: F401 + ) + from sklearn.utils.multiclass import type_of_target # noqa: F401 + + # validation + from sklearn.utils.validation import ( + _check_feature_names, # noqa: F401 + _check_n_features, # noqa: F401 + validate_data, # noqa: F401 + ) From 2dc7ebc177fc7604c3475b3ecc14298c9fef4d59 Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 22:26:32 +0100 Subject: [PATCH 02/10] compat python 3.7 --- python-package/lightgbm/sklearn_compat.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python-package/lightgbm/sklearn_compat.py b/python-package/lightgbm/sklearn_compat.py index 9ce6792203ce..e79ce2511900 100644 --- a/python-package/lightgbm/sklearn_compat.py +++ b/python-package/lightgbm/sklearn_compat.py @@ -325,7 +325,7 @@ def _construct_instances(Estimator): yield _construct_instance(Estimator) # validation - def validate_data(_estimator, /, **kwargs): + def validate_data(_estimator, **kwargs): if "ensure_all_finite" in kwargs: force_all_finite = kwargs.pop("ensure_all_finite") else: From 40952803b7d9e51ba7315eed085270f71b3becab Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 22:34:24 +0100 Subject: [PATCH 03/10] fixing lint --- python-package/lightgbm/compat.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python-package/lightgbm/compat.py b/python-package/lightgbm/compat.py index 9a3f3a248b6a..c37c0207110a 100644 --- a/python-package/lightgbm/compat.py +++ b/python-package/lightgbm/compat.py @@ -64,7 +64,7 @@ class _LGBMRegressorBase: # type: ignore # additional scikit-learn imports only for type hints if TYPE_CHECKING: - from .sklearn_compat.utils import Tags as _sklearn_Tags + from .sklearn_compat.utils import Tags as _sklearn_Tags # noqa: F401 """pandas""" try: From 2585bdd623d15ff42233891b5b28ea72da8f706f Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 22:34:40 +0100 Subject: [PATCH 04/10] fixing lint --- python-package/pyproject.toml | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/python-package/pyproject.toml b/python-package/pyproject.toml index 19866e01202b..76e28bdf61df 100644 --- a/python-package/pyproject.toml +++ b/python-package/pyproject.toml @@ -185,6 +185,13 @@ select = [ # (pylint) Using the global statement is discouraged "PLW0603" ] +"python-package/lightgbm/sklearn_compat.py" = [ + # file is vendored from sklearn-compat + "D401", + "D400", + "D103", + "D205", +] "tests/*" = [ # (flake8-bugbear) Found useless expression "B018", From 3c2f26f6af679a06cf6f99508ead452aa6b8ceda Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 22:40:15 +0100 Subject: [PATCH 05/10] debug --- python-package/lightgbm/compat.py | 1 + 1 file changed, 1 insertion(+) diff --git a/python-package/lightgbm/compat.py b/python-package/lightgbm/compat.py index c37c0207110a..d0420ed9a543 100644 --- a/python-package/lightgbm/compat.py +++ b/python-package/lightgbm/compat.py @@ -33,6 +33,7 @@ _LGBMComputeSampleWeight = compute_sample_weight _LGBMValidateData = validate_data except ImportError: + raise SKLEARN_INSTALLED = False class _LGBMModelBase: # type: ignore From 37f209ee6455f2bff9cc2ca558747b5c4532e1e5 Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 22:47:42 +0100 Subject: [PATCH 06/10] use vendoring import --- .../lightgbm/{sklearn_compat.py => _sklearn_compat.py} | 1 + python-package/lightgbm/compat.py | 4 ++-- 2 files changed, 3 insertions(+), 2 deletions(-) rename python-package/lightgbm/{sklearn_compat.py => _sklearn_compat.py} (99%) diff --git a/python-package/lightgbm/sklearn_compat.py b/python-package/lightgbm/_sklearn_compat.py similarity index 99% rename from python-package/lightgbm/sklearn_compat.py rename to python-package/lightgbm/_sklearn_compat.py index e79ce2511900..c1414f1daf5d 100644 --- a/python-package/lightgbm/sklearn_compat.py +++ b/python-package/lightgbm/_sklearn_compat.py @@ -1,3 +1,4 @@ +# coding: utf-8 """Ease developer experience to support multiple versions of scikit-learn. This file is intended to be vendored in your project if you do not want to depend on diff --git a/python-package/lightgbm/compat.py b/python-package/lightgbm/compat.py index d0420ed9a543..45650a93f1a5 100644 --- a/python-package/lightgbm/compat.py +++ b/python-package/lightgbm/compat.py @@ -16,7 +16,7 @@ from sklearn.exceptions import NotFittedError from sklearn.model_selection import BaseCrossValidator, GroupKFold, StratifiedKFold from sklearn.utils.validation import _check_sample_weight - from .sklearn_compat.utils.validation import validate_data + from ._sklearn_compat import validate_data SKLEARN_INSTALLED = True _LGBMBaseCrossValidator = BaseCrossValidator @@ -65,7 +65,7 @@ class _LGBMRegressorBase: # type: ignore # additional scikit-learn imports only for type hints if TYPE_CHECKING: - from .sklearn_compat.utils import Tags as _sklearn_Tags # noqa: F401 + from ._sklearn_compat import Tags as _sklearn_Tags # noqa: F401 """pandas""" try: From 2a658a8f34e725fa81402420f516f995976842f7 Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 22:48:11 +0100 Subject: [PATCH 07/10] remove debug --- python-package/lightgbm/compat.py | 1 - 1 file changed, 1 deletion(-) diff --git a/python-package/lightgbm/compat.py b/python-package/lightgbm/compat.py index 45650a93f1a5..377d1fae47c5 100644 --- a/python-package/lightgbm/compat.py +++ b/python-package/lightgbm/compat.py @@ -33,7 +33,6 @@ _LGBMComputeSampleWeight = compute_sample_weight _LGBMValidateData = validate_data except ImportError: - raise SKLEARN_INSTALLED = False class _LGBMModelBase: # type: ignore From 806b9fd22d90b0a73ebd83d9b211083e90213a1f Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 22:55:51 +0100 Subject: [PATCH 08/10] fix --- python-package/lightgbm/sklearn.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/python-package/lightgbm/sklearn.py b/python-package/lightgbm/sklearn.py index 108ef1e14498..bdb9224a3bdf 100644 --- a/python-package/lightgbm/sklearn.py +++ b/python-package/lightgbm/sklearn.py @@ -919,8 +919,8 @@ def fit( if not isinstance(X, (pd_DataFrame, dt_DataTable)): _X, _y = _LGBMValidateData( self, - X, - y, + X=X, + y=y, reset=True, # allow any input type (this validation is done further down, in lgb.Dataset()) accept_sparse=True, @@ -1078,7 +1078,7 @@ def predict( if not isinstance(X, (pd_DataFrame, dt_DataTable)): X = _LGBMValidateData( self, - X, + X=X, # 'y' being omitted = run scikit-learn's check_array() instead of check_X_y() # # Prevent scikit-learn from deleting or modifying attributes like 'feature_names_in_' and 'n_features_in_'. From 879aea68b0e9b557d7d817383d3507e44f6442a3 Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 23:05:04 +0100 Subject: [PATCH 09/10] debug --- python-package/lightgbm/compat.py | 1 + 1 file changed, 1 insertion(+) diff --git a/python-package/lightgbm/compat.py b/python-package/lightgbm/compat.py index 377d1fae47c5..45650a93f1a5 100644 --- a/python-package/lightgbm/compat.py +++ b/python-package/lightgbm/compat.py @@ -33,6 +33,7 @@ _LGBMComputeSampleWeight = compute_sample_weight _LGBMValidateData = validate_data except ImportError: + raise SKLEARN_INSTALLED = False class _LGBMModelBase: # type: ignore From 4e4ca84c64c7a7b53ea632fc033b5cba16ed9eb2 Mon Sep 17 00:00:00 2001 From: Guillaume Lemaitre Date: Sun, 8 Dec 2024 23:15:22 +0100 Subject: [PATCH 10/10] remove type annotation --- python-package/lightgbm/_sklearn_compat.py | 15 +++++++-------- 1 file changed, 7 insertions(+), 8 deletions(-) diff --git a/python-package/lightgbm/_sklearn_compat.py b/python-package/lightgbm/_sklearn_compat.py index c1414f1daf5d..1a6da7982102 100644 --- a/python-package/lightgbm/_sklearn_compat.py +++ b/python-package/lightgbm/_sklearn_compat.py @@ -15,7 +15,6 @@ import platform import sys from dataclasses import dataclass, field -from typing import Callable, Literal import sklearn from sklearn.utils._param_validation import validate_parameter_constraints @@ -586,11 +585,11 @@ def check_estimator( estimator=None, generate_only=False, *, - legacy: bool = True, - expected_failed_checks: dict[str, str] | None = None, - on_skip: Literal["warn"] | None = "warn", - on_fail: Literal["raise", "warn"] | None = "raise", - callback: Callable | None = None, + legacy=True, + expected_failed_checks=None, + on_skip="warn", + on_fail="raise", + callback=None, ): # legacy, on_skip, on_fail, and callback are not supported and ignored from sklearn.utils.estimator_checks import check_estimator @@ -603,8 +602,8 @@ def check_estimator( def parametrize_with_checks( estimators, *, - legacy: bool = True, - expected_failed_checks: Callable | None = None, + legacy=True, + expected_failed_checks=None, ): # legacy is not supported and ignored from sklearn.utils.estimator_checks import parametrize_with_checks