hyperactive.experiment.integrations.sklearn_cv.SklearnCvExperiment._evaluate() - Code Metrics - Inspection of "[ENH] sign handling in experiments and optimizatio..." - SimonBlanke/Hyperactive - Measure and Improve Code Quality continuously with Scrutinizer

Passed

Push — master ( 4a40b1...18f8a0 )

by Simon

created 2025-07-28 16:07 UTC

SklearnCvExperiment._evaluate() A

↳ Parent: hyperactive.experiment.integrations.sklearn_cv

Complexity

Conditions

Size

Total Lines	33
Code Lines	14

Duplication

Lines	0
Ratio	0 %

Importance

Changes

Metric	Value
eloc	14
dl	0
loc	33
rs	9.7
c	0
b	0
f	0
cc	1
nop	2

"""Experiment adapter for sklearn cross-validation experiments."""
# copyright: hyperactive developers, MIT License (see LICENSE file)

from sklearn import clone
from sklearn.metrics import check_scoring
from sklearn.model_selection import cross_validate
from sklearn.utils.validation import _num_samples

from hyperactive.base import BaseExperiment

class SklearnCvExperiment(BaseExperiment):
    """Experiment adapter for sklearn cross-validation experiments.

    This class is used to perform cross-validation experiments using a given
    sklearn estimator. It allows for hyperparameter tuning and evaluation of
    the model's performance using cross-validation.

    The score returned is the mean of the cross-validation scores,
    of applying cross-validation to ``estimator`` with the parameters given in
    ``score`` ``params``.

    The cross-validation performed is specified by the ``cv`` parameter,
    and the scoring metric is specified by the ``scoring`` parameter.
    The ``X`` and ``y`` parameters are the input data and target values,
    which are used in fit/predict cross-validation.

    Parameters
    ----------
    estimator : sklearn estimator
        The estimator to be used for the experiment.
    X : array-like, shape (n_samples, n_features)
            The input data for the model.
    y : array-like, shape (n_samples,) or (n_samples, n_outputs)
        The target values for the model.
    cv : int or cross-validation generator, default = KFold(n_splits=3, shuffle=True)
        The number of folds or cross-validation strategy to be used.
        If int, the cross-validation used is KFold(n_splits=cv, shuffle=True).
    scoring : callable or str, default = accuracy_score or mean_squared_error
        sklearn scoring function or metric to evaluate the model's performance.
        Default is determined by the type of estimator:
        ``accuracy_score`` for classifiers, and
        ``mean_squared_error`` for regressors, as per sklearn convention
        through the default ``score`` method of the estimator.

    Example
    -------
    >>> from hyperactive.experiment.integrations import SklearnCvExperiment
    >>> from sklearn.datasets import load_iris
    >>> from sklearn.svm import SVC
    >>> from sklearn.metrics import accuracy_score
    >>> from sklearn.model_selection import KFold
    >>>
    >>> X, y = load_iris(return_X_y=True)
    >>>
    >>> sklearn_exp = SklearnCvExperiment(
    ...     estimator=SVC(),
    ...     scoring=accuracy_score,
    ...     cv=KFold(n_splits=3, shuffle=True),
    ...     X=X,
    ...     y=y,
    ... )
    >>> params = {"C": 1.0, "kernel": "linear"}
    >>> score, add_info = sklearn_exp.score(params)

    For default choices of ``scoring`` and ``cv``:
    >>> sklearn_exp = SklearnCvExperiment(
    ...     estimator=SVC(),
    ...     X=X,
    ...     y=y,
    ... )
    >>> params = {"C": 1.0, "kernel": "linear"}
    >>> score, add_info = sklearn_exp.score(params)

    Quick call without metadata return or dictionary:
    >>> score = sklearn_exp(C=1.0, kernel="linear")
    """

    def __init__(self, estimator, X, y, scoring=None, cv=None):
        self.estimator = estimator
        self.X = X
        self.y = y
        self.scoring = scoring
        self.cv = cv

        super().__init__()

        if cv is None:
            from sklearn.model_selection import KFold

            self._cv = KFold(n_splits=3, shuffle=True)
        elif isinstance(cv, int):
            from sklearn.model_selection import KFold

            self._cv = KFold(n_splits=cv, shuffle=True)
        else:
            self._cv = cv

        # check if scoring is a scorer by checking for "estimator" in signature
        if scoring is None:
            self._scoring = check_scoring(self.estimator)
        # check using inspect.signature for "estimator" in signature
        elif callable(scoring):
            from inspect import signature

            if "estimator" in signature(scoring).parameters:
                self._scoring = scoring
            else:
                from sklearn.metrics import make_scorer

                self._scoring = make_scorer(scoring)
        self.scorer_ = self._scoring

        # Set the sign of the scoring function
        if hasattr(self._scoring, "_score"):
            score_func = self._scoring._score_func
            _sign = _guess_sign_of_sklmetric(score_func)
            _sign_str = "higher" if _sign == 1 else "lower"
            self.set_tags(**{"property:higher_or_lower_is_better": _sign_str})

    def _paramnames(self):
        """Return the parameter names of the search.

        Returns
        -------
        list of str
            The parameter names of the search parameters.
        """
        return list(self.estimator.get_params().keys())

    def _evaluate(self, params):
        """Evaluate the parameters.

        Parameters
        ----------
        params : dict with string keys
            Parameters to evaluate.

        Returns
        -------
        float
            The value of the parameters as per evaluation.
        dict
            Additional metadata about the search.
        """
        estimator = clone(self.estimator)
        estimator.set_params(**params)

        cv_results = cross_validate(
            estimator,
            self.X,
            self.y,
            scoring=self._scoring,
            cv=self._cv,
        )

        add_info_d = {
            "score_time": cv_results["score_time"],
            "fit_time": cv_results["fit_time"],
            "n_test_samples": _num_samples(self.X),
        }

        return cv_results["test_score"].mean(), add_info_d

    @classmethod
    def get_test_params(cls, parameter_set="default"):
        """Return testing parameter settings for the skbase object.

        ``get_test_params`` is a unified interface point to store
        parameter settings for testing purposes. This function is also
        used in ``create_test_instance`` and ``create_test_instances_and_names``
        to construct test instances.

        ``get_test_params`` should return a single ``dict``, or a ``list`` of ``dict``.

        Each ``dict`` is a parameter configuration for testing,
        and can be used to construct an "interesting" test instance.
        A call to ``cls(**params)`` should
        be valid for all dictionaries ``params`` in the return of ``get_test_params``.

        The ``get_test_params`` need not return fixed lists of dictionaries,
        it can also return dynamic or stochastic parameter settings.

        Parameters
        ----------
        parameter_set : str, default="default"
            Name of the set of test parameters to return, for use in tests. If no
            special parameters are defined for a value, will return `"default"` set.

        Returns
        -------
        params : dict or list of dict, default = {}
            Parameters to create testing instances of the class
            Each dict are parameters to construct an "interesting" test instance, i.e.,
            `MyClass(**params)` or `MyClass(**params[i])` creates a valid test instance.
            `create_test_instance` uses the first (or only) dictionary in `params`
        """
        from sklearn.datasets import load_diabetes, load_iris
        from sklearn.svm import SVC, SVR
        from sklearn.metrics import accuracy_score, mean_absolute_error
        from sklearn.model_selection import KFold

        X, y = load_iris(return_X_y=True)
        params_classif = {
            "estimator": SVC(),
            "scoring": accuracy_score,
            "cv": KFold(n_splits=3, shuffle=True),
            "X": X,
            "y": y,
        }

        X, y = load_diabetes(return_X_y=True)
        params_regress = {
            "estimator": SVR(),
            "scoring": mean_absolute_error,
            "cv": 2,
            "X": X,
            "y": y,
        }

        X, y = load_diabetes(return_X_y=True)
        params_all_default = {
            "estimator": SVR(),
            "X": X,
            "y": y,
        }

        return [params_classif, params_regress, params_all_default]

    @classmethod
    def _get_score_params(self):
        """Return settings for testing score/evaluate functions. Used in tests only.

        Returns a list, the i-th element should be valid arguments for
        self.evaluate and self.score, of an instance constructed with
        self.get_test_params()[i].

        Returns
        -------
        list of dict
            The parameters to be used for scoring.
        """
        score_params_classif = {"C": 1.0, "kernel": "linear"}
        score_params_regress = {"C": 1.0, "kernel": "linear"}
        score_params_defaults = {"C": 1.0, "kernel": "linear"}
        return [score_params_classif, score_params_regress, score_params_defaults]


def _guess_sign_of_sklmetric(scorer):
    """Guess the sign of a sklearn metric scorer.

    Parameters
    ----------
    scorer : callable
        The sklearn metric scorer to guess the sign for.

    Returns
    -------
    int
        1 if higher scores are better, -1 if lower scores are better.
    """
    HIGHER_IS_BETTER = {
        # Classification
        "accuracy_score": True,
        "auc": True,
        "average_precision_score": True,
        "balanced_accuracy_score": True,
        "brier_score_loss": False,
        "class_likelihood_ratios": False,
        "cohen_kappa_score": True,
        "d2_log_loss_score": True,
        "dcg_score": True,
        "f1_score": True,
        "fbeta_score": True,
        "hamming_loss": False,
        "hinge_loss": False,
        "jaccard_score": True,
        "log_loss": False,
        "matthews_corrcoef": True,
        "ndcg_score": True,
        "precision_score": True,
        "recall_score": True,
        "roc_auc_score": True,
        "top_k_accuracy_score": True,
        "zero_one_loss": False,

        # Regression
        "d2_absolute_error_score": True,
        "d2_pinball_score": True,
        "d2_tweedie_score": True,
        "explained_variance_score": True,
        "max_error": False,
        "mean_absolute_error": False,
        "mean_absolute_percentage_error": False,
        "mean_gamma_deviance": False,
        "mean_pinball_loss": False,
        "mean_poisson_deviance": False,
        "mean_squared_error": False,
        "mean_squared_log_error": False,
        "mean_tweedie_deviance": False,
        "median_absolute_error": False,
        "r2_score": True,
        "root_mean_squared_error": False,
        "root_mean_squared_log_error": False,
    }

    scorer_name = getattr(scorer, "__name__", None)

    if hasattr(scorer, "greater_is_better"):
        return 1 if scorer.greater_is_better else -1
    elif scorer_name in HIGHER_IS_BETTER:
        return 1 if HIGHER_IS_BETTER[scorer_name] else -1
    elif scorer_name.endswith("_score"):
        # If the scorer name ends with "_score", we assume higher is better
        return 1
    elif scorer_name.endswith("_loss") or scorer_name.endswith("_deviance"):
        # If the scorer name ends with "_loss", we assume lower is better
        return -1
    elif scorer_name.endswith("_error"):
        return -1
    else:
        # If we cannot determine the sign, we assume lower is better
        return -1


1			"""Experiment adapter for sklearn cross-validation experiments."""
2			# copyright: hyperactive developers, MIT License (see LICENSE file)
3
4			from sklearn import clone
5			from sklearn.metrics import check_scoring
6			from sklearn.model_selection import cross_validate
7			from sklearn.utils.validation import _num_samples
8
9			from hyperactive.base import BaseExperiment
10
11			class SklearnCvExperiment(BaseExperiment):
12			"""Experiment adapter for sklearn cross-validation experiments.
13
14			This class is used to perform cross-validation experiments using a given
15			sklearn estimator. It allows for hyperparameter tuning and evaluation of
16			the model's performance using cross-validation.
17
18			The score returned is the mean of the cross-validation scores,
19			of applying cross-validation to ``estimator`` with the parameters given in
20			``score`` ``params``.
21
22			The cross-validation performed is specified by the ``cv`` parameter,
23			and the scoring metric is specified by the ``scoring`` parameter.
24			The ``X`` and ``y`` parameters are the input data and target values,
25			which are used in fit/predict cross-validation.
26
27			Parameters
28			----------
29			estimator : sklearn estimator
30			The estimator to be used for the experiment.
31			X : array-like, shape (n_samples, n_features)
32			The input data for the model.
33			y : array-like, shape (n_samples,) or (n_samples, n_outputs)
34			The target values for the model.
35			cv : int or cross-validation generator, default = KFold(n_splits=3, shuffle=True)
36			The number of folds or cross-validation strategy to be used.
37			If int, the cross-validation used is KFold(n_splits=cv, shuffle=True).
38			scoring : callable or str, default = accuracy_score or mean_squared_error
39			sklearn scoring function or metric to evaluate the model's performance.
40			Default is determined by the type of estimator:
41			``accuracy_score`` for classifiers, and
42			``mean_squared_error`` for regressors, as per sklearn convention
43			through the default ``score`` method of the estimator.
44
45			Example
46			-------
47			>>> from hyperactive.experiment.integrations import SklearnCvExperiment
48			>>> from sklearn.datasets import load_iris
49			>>> from sklearn.svm import SVC
50			>>> from sklearn.metrics import accuracy_score
51			>>> from sklearn.model_selection import KFold
52			>>>
53			>>> X, y = load_iris(return_X_y=True)
54			>>>
55			>>> sklearn_exp = SklearnCvExperiment(
56			... estimator=SVC(),
57			... scoring=accuracy_score,
58			... cv=KFold(n_splits=3, shuffle=True),
59			... X=X,
60			... y=y,
61			... )
62			>>> params = {"C": 1.0, "kernel": "linear"}
63			>>> score, add_info = sklearn_exp.score(params)
64
65			For default choices of ``scoring`` and ``cv``:
66			>>> sklearn_exp = SklearnCvExperiment(
67			... estimator=SVC(),
68			... X=X,
69			... y=y,
70			... )
71			>>> params = {"C": 1.0, "kernel": "linear"}
72			>>> score, add_info = sklearn_exp.score(params)
73
74			Quick call without metadata return or dictionary:
75			>>> score = sklearn_exp(C=1.0, kernel="linear")
76			"""
77
78			def __init__(self, estimator, X, y, scoring=None, cv=None):
79			self.estimator = estimator
80			self.X = X
81			self.y = y
82			self.scoring = scoring
83			self.cv = cv
84
85			super().__init__()
86
87			if cv is None:
88			from sklearn.model_selection import KFold
89
90			self._cv = KFold(n_splits=3, shuffle=True)
91			elif isinstance(cv, int):
92			from sklearn.model_selection import KFold
93
94			self._cv = KFold(n_splits=cv, shuffle=True)
95			else:
96			self._cv = cv
97
98			# check if scoring is a scorer by checking for "estimator" in signature
99			if scoring is None:
100			self._scoring = check_scoring(self.estimator)
101			# check using inspect.signature for "estimator" in signature
102			elif callable(scoring):
103			from inspect import signature
104
105			if "estimator" in signature(scoring).parameters:
106			self._scoring = scoring
107			else:
108			from sklearn.metrics import make_scorer
109
110			self._scoring = make_scorer(scoring)
111			self.scorer_ = self._scoring
112
113			# Set the sign of the scoring function
114			if hasattr(self._scoring, "_score"):
115			score_func = self._scoring._score_func
116			_sign = _guess_sign_of_sklmetric(score_func)
117			_sign_str = "higher" if _sign == 1 else "lower"
118			self.set_tags(**{"property:higher_or_lower_is_better": _sign_str})
119
120			def _paramnames(self):
121			"""Return the parameter names of the search.
122
123			Returns
124			-------
125			list of str
126			The parameter names of the search parameters.
127			"""
128			return list(self.estimator.get_params().keys())
129
130			def _evaluate(self, params):
131			"""Evaluate the parameters.
132
133			Parameters
134			----------
135			params : dict with string keys
136			Parameters to evaluate.
137
138			Returns
139			-------
140			float
141			The value of the parameters as per evaluation.
142			dict
143			Additional metadata about the search.
144			"""
145			estimator = clone(self.estimator)
146			estimator.set_params(**params)
147
148			cv_results = cross_validate(
149			estimator,
150			self.X,
151			self.y,
152			scoring=self._scoring,
153			cv=self._cv,
154			)
155
156			add_info_d = {
157			"score_time": cv_results["score_time"],
158			"fit_time": cv_results["fit_time"],
159			"n_test_samples": _num_samples(self.X),
160			}
161
162			return cv_results["test_score"].mean(), add_info_d
163
164			@classmethod
165			def get_test_params(cls, parameter_set="default"):
166			"""Return testing parameter settings for the skbase object.
167
168			``get_test_params`` is a unified interface point to store
169			parameter settings for testing purposes. This function is also
170			used in ``create_test_instance`` and ``create_test_instances_and_names``
171			to construct test instances.
172
173			``get_test_params`` should return a single ``dict``, or a ``list`` of ``dict``.
174
175			Each ``dict`` is a parameter configuration for testing,
176			and can be used to construct an "interesting" test instance.
177			A call to ``cls(**params)`` should
178			be valid for all dictionaries ``params`` in the return of ``get_test_params``.
179
180			The ``get_test_params`` need not return fixed lists of dictionaries,
181			it can also return dynamic or stochastic parameter settings.
182
183			Parameters
184			----------
185			parameter_set : str, default="default"
186			Name of the set of test parameters to return, for use in tests. If no
187			special parameters are defined for a value, will return `"default"` set.
188
189			Returns
190			-------
191			params : dict or list of dict, default = {}
192			Parameters to create testing instances of the class
193			Each dict are parameters to construct an "interesting" test instance, i.e.,
194			`MyClass(params)` or `MyClass(params[i])` creates a valid test instance.
195			`create_test_instance` uses the first (or only) dictionary in `params`
196			"""
197			from sklearn.datasets import load_diabetes, load_iris
198			from sklearn.svm import SVC, SVR
199			from sklearn.metrics import accuracy_score, mean_absolute_error
200			from sklearn.model_selection import KFold
201
202			X, y = load_iris(return_X_y=True)
203			params_classif = {
204			"estimator": SVC(),
205			"scoring": accuracy_score,
206			"cv": KFold(n_splits=3, shuffle=True),
207			"X": X,
208			"y": y,
209			}
210
211			X, y = load_diabetes(return_X_y=True)
212			params_regress = {
213			"estimator": SVR(),
214			"scoring": mean_absolute_error,
215			"cv": 2,
216			"X": X,
217			"y": y,
218			}
219
220			X, y = load_diabetes(return_X_y=True)
221			params_all_default = {
222			"estimator": SVR(),
223			"X": X,
224			"y": y,
225			}
226
227			return [params_classif, params_regress, params_all_default]
228
229			@classmethod
230			def _get_score_params(self):
231			"""Return settings for testing score/evaluate functions. Used in tests only.
232
233			Returns a list, the i-th element should be valid arguments for
234			self.evaluate and self.score, of an instance constructed with
235			self.get_test_params()[i].
236
237			Returns
238			-------
239			list of dict
240			The parameters to be used for scoring.
241			"""
242			score_params_classif = {"C": 1.0, "kernel": "linear"}
243			score_params_regress = {"C": 1.0, "kernel": "linear"}
244			score_params_defaults = {"C": 1.0, "kernel": "linear"}
245			return [score_params_classif, score_params_regress, score_params_defaults]
246
247
248			def _guess_sign_of_sklmetric(scorer):
249			"""Guess the sign of a sklearn metric scorer.
250
251			Parameters
252			----------
253			scorer : callable
254			The sklearn metric scorer to guess the sign for.
255
256			Returns
257			-------
258			int
259			1 if higher scores are better, -1 if lower scores are better.
260			"""
261			HIGHER_IS_BETTER = {
262			# Classification
263			"accuracy_score": True,
264			"auc": True,
265			"average_precision_score": True,
266			"balanced_accuracy_score": True,
267			"brier_score_loss": False,
268			"class_likelihood_ratios": False,
269			"cohen_kappa_score": True,
270			"d2_log_loss_score": True,
271			"dcg_score": True,
272			"f1_score": True,
273			"fbeta_score": True,
274			"hamming_loss": False,
275			"hinge_loss": False,
276			"jaccard_score": True,
277			"log_loss": False,
278			"matthews_corrcoef": True,
279			"ndcg_score": True,
280			"precision_score": True,
281			"recall_score": True,
282			"roc_auc_score": True,
283			"top_k_accuracy_score": True,
284			"zero_one_loss": False,
285
286			# Regression
287			"d2_absolute_error_score": True,
288			"d2_pinball_score": True,
289			"d2_tweedie_score": True,
290			"explained_variance_score": True,
291			"max_error": False,
292			"mean_absolute_error": False,
293			"mean_absolute_percentage_error": False,
294			"mean_gamma_deviance": False,
295			"mean_pinball_loss": False,
296			"mean_poisson_deviance": False,
297			"mean_squared_error": False,
298			"mean_squared_log_error": False,
299			"mean_tweedie_deviance": False,
300			"median_absolute_error": False,
301			"r2_score": True,
302			"root_mean_squared_error": False,
303			"root_mean_squared_log_error": False,
304			}
305
306			scorer_name = getattr(scorer, "__name__", None)
307
308			if hasattr(scorer, "greater_is_better"):
309			return 1 if scorer.greater_is_better else -1
310			elif scorer_name in HIGHER_IS_BETTER:
311			return 1 if HIGHER_IS_BETTER[scorer_name] else -1
312			elif scorer_name.endswith("_score"):
313			# If the scorer name ends with "_score", we assume higher is better
314			return 1
315			elif scorer_name.endswith("_loss") or scorer_name.endswith("_deviance"):
316			# If the scorer name ends with "_loss", we assume lower is better
317			return -1
318			elif scorer_name.endswith("_error"):
319			return -1
320			else:
321			# If we cannot determine the sign, we assume lower is better
322			return -1
323

SimonBlanke / Hyperactive

Push — master ( 4a40b1...18f8a0 )

SklearnCvExperiment._evaluate() A

Complexity

Size

Duplication

Importance

Duplication Side-by-Side

Filter issues like