TunedThresholdClassifierCV

classsklearn.model_selection.TunedThresholdClassifierCV(estimator, *, scoring='balanced_accuracy', response_method='auto', thresholds=100, cv=None, refit=True, n_jobs=None, random_state=None, store_cv_results=False)[source]

Classifier that post-tunes the decision threshold using cross-validation.

This estimator post-tunes the decision threshold (cut-off point) that is used for converting posterior probability estimates (i.e. output of predict_proba) or decision scores (i.e. output of decision_function) into a class label. The tuning is done by optimizing a binary metric, potentially constrained by another metric.

Read more in the User Guide.

Added in version 1.5.

Parameters:

estimatorestimator instance

The classifier, fitted or not, for which we want to optimize the decision threshold used during predict.

scoringstr or callable, default=”balanced_accuracy”

The objective metric to be optimized. Can be one of:

• str: string associated to a scoring function for binary classification, see String name scorers for options.
• callable: a scorer callable object (e.g., function) with signature scorer(estimator, X, y). See Callable scorers for details.

response_method{“auto”, “decision_function”, “predict_proba”}, default=”auto”

Methods by the classifier estimator corresponding to the decision function for which we want to find a threshold. It can be:

• if "auto", it will try to invoke, for each classifier, "predict_proba" or "decision_function" in that order.
• otherwise, one of "predict_proba" or "decision_function". If the method is not implemented by the classifier, it will raise an error.

thresholdsint or array-like, default=100

The number of decision threshold to use when discretizing the output of the classifier method. Pass an array-like to manually specify the thresholds to use.

cvint, float, cross-validation generator, iterable or “prefit”, default=None

Determines the cross-validation splitting strategy to train classifier. Possible inputs for cv are:

• None, to use the default 5-fold stratified K-fold cross validation;
• An integer number, to specify the number of folds in a stratified k-fold;
• A float number, to specify a single shuffle split. The floating number should be in (0, 1) and represent the size of the validation set;
• An object to be used as a cross-validation generator;
• An iterable yielding train, test splits;
• "prefit", to bypass the cross-validation.

Refer User Guide for the various cross-validation strategies that can be used here.

Warning

Using cv="prefit" and passing the same dataset for fitting estimator and tuning the cut-off point is subject to undesired overfitting. You can refer to Consideration regarding model refitting and cross-validation for an example.

This option should only be used when the set used to fit estimator is different from the one used to tune the cut-off point (by calling TunedThresholdClassifierCV.fit).

refitbool, default=True

Whether or not to refit the classifier on the entire training set once the decision threshold has been found. Note that forcing refit=False on cross-validation having more than a single split will raise an error. Similarly, refit=True in conjunction with cv="prefit" will raise an error.

n_jobsint, default=None

The number of jobs to run in parallel. When cv represents a cross-validation strategy, the fitting and scoring on each data split is done in parallel. None means 1 unless in a joblib.parallel_backend context. -1 means using all processors. See Glossary for more details.

random_stateint, RandomState instance or None, default=None

Controls the randomness of cross-validation when cv is a float. See Glossary.

store_cv_resultsbool, default=False

Whether to store all scores and thresholds computed during the cross-validation process.

Attributes:

estimator_estimator instance

The fitted classifier used when predicting.

best_threshold_float

The new decision threshold.

best_score_float or None

The optimal score of the objective metric, evaluated at best_threshold_.

cv_results_dict or None

A dictionary containing the scores and thresholds computed during the cross-validation process. Only exist if store_cv_results=True. The keys are "thresholds" and "scores".

classes_ndarray of shape (n_classes,)

Classes labels.

n_features_in_int

Number of features seen during fit. Only defined if the underlying estimator exposes such an attribute when fit.

feature_names_in_ndarray of shape (n_features_in_,)

Names of features seen during fit. Only defined if the underlying estimator exposes such an attribute when fit.

See also

sklearn.model_selection.FixedThresholdClassifier

Classifier that uses a constant threshold.

sklearn.calibration.CalibratedClassifierCV

Estimator that calibrates probabilities.

Examples

>>> from sklearn.datasets import make_classification
>>> from sklearn.ensemble import RandomForestClassifier
>>> from sklearn.metrics import classification_report
>>> from sklearn.model_selection import TunedThresholdClassifierCV, train_test_split
>>> X, y = make_classification(
...     n_samples=1_000, weights=[0.9, 0.1], class_sep=0.8, random_state=42
... )
>>> X_train, X_test, y_train, y_test = train_test_split(
...     X, y, stratify=y, random_state=42
... )
>>> classifier = RandomForestClassifier(random_state=0).fit(X_train, y_train)
>>> print(classification_report(y_test, classifier.predict(X_test)))
              precision    recall  f1-score   support

           0       0.94      0.99      0.96       224
           1       0.80      0.46      0.59        26

    accuracy                           0.93       250
   macro avg       0.87      0.72      0.77       250
weighted avg       0.93      0.93      0.92       250

>>> classifier_tuned = TunedThresholdClassifierCV(
...     classifier, scoring="balanced_accuracy"
... ).fit(X_train, y_train)
>>> print(
...     f"Cut-off point found at {classifier_tuned.best_threshold_:.3f}"
... )
Cut-off point found at 0.342
>>> print(classification_report(y_test, classifier_tuned.predict(X_test)))
              precision    recall  f1-score   support

           0       0.96      0.95      0.96       224
           1       0.61      0.65      0.63        26

    accuracy                           0.92       250
   macro avg       0.78      0.80      0.79       250
weighted avg       0.92      0.92      0.92       250

decision_function(X)[source]

Decision function for samples in X using the fitted estimator.

Parameters:

X{array-like, sparse matrix} of shape (n_samples, n_features)

Training vectors, where n_samples is the number of samples and n_features is the number of features.

Returns:

decisionsndarray of shape (n_samples,)

The decision function computed the fitted estimator.

fit(X, y, **params)[source]

Fit the classifier.

Parameters:

X{array-like, sparse matrix} of shape (n_samples, n_features)

Training data.

yarray-like of shape (n_samples,)

Target values.

**paramsdict

Parameters to pass to the fit method of the underlying classifier.

Returns:

selfobject

Returns an instance of self.

get_metadata_routing()[source]

Get metadata routing of this object.

Please check User Guide on how the routing mechanism works.

Returns:

routingMetadataRouter

A MetadataRouter encapsulating routing information.

get_params(deep=True)[source]

Get parameters for this estimator.

Parameters:

deepbool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns:

paramsdict

Parameter names mapped to their values.

predict(X)[source]

Predict the target of new samples.

Parameters:

X{array-like, sparse matrix} of shape (n_samples, n_features)

The samples, as accepted by estimator.predict.

Returns:

class_labelsndarray of shape (n_samples,)

The predicted class.

predict_log_proba(X)[source]

Predict logarithm class probabilities for X using the fitted estimator.

Parameters:

X{array-like, sparse matrix} of shape (n_samples, n_features)

Training vectors, where n_samples is the number of samples and n_features is the number of features.

Returns:

log_probabilitiesndarray of shape (n_samples, n_classes)

The logarithm class probabilities of the input samples.

predict_proba(X)[source]

Predict class probabilities for X using the fitted estimator.

Parameters:

X{array-like, sparse matrix} of shape (n_samples, n_features)

Training vectors, where n_samples is the number of samples and n_features is the number of features.

Returns:

probabilitiesndarray of shape (n_samples, n_classes)

The class probabilities of the input samples.

score(X, y, sample_weight=None)[source]

Return accuracy on provided data and labels.

In multi-label classification, this is the subset accuracy which is a harsh metric since you require for each sample that each label set be correctly predicted.

Parameters:

Xarray-like of shape (n_samples, n_features)

Test samples.

yarray-like of shape (n_samples,) or (n_samples, n_outputs)

True labels for X.

sample_weightarray-like of shape (n_samples,), default=None

Sample weights.

Returns:

scorefloat

Mean accuracy of self.predict(X) w.r.t. y.

set_params(**params)[source]

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as Pipeline). The latter have parameters of the form <component>__<parameter> so that it’s possible to update each component of a nested object.

Parameters:

**paramsdict

Estimator parameters.

Returns:

selfestimator instance

Estimator instance.

set_score_request(*, sample_weight:bool|None|str='$UNCHANGED$') → TunedThresholdClassifierCV[source]

Configure whether metadata should be requested to be passed to the score method.

Note that this method is only relevant when this estimator is used as a sub-estimator within a meta-estimator and metadata routing is enabled with enable_metadata_routing=True (see sklearn.set_config). Please check the User Guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to score if provided. The request is ignored if metadata is not provided.
• False: metadata is not requested and the meta-estimator will not pass it to score.
• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.
• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others.

Added in version 1.3.

Parameters:

sample_weightstr, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for sample_weight parameter in score.

Returns:

selfobject

The updated object.

Gallery examples

Post-tuning the decision threshold for cost-sensitive learning

Post-hoc tuning the cut-off point of decision function

Release Highlights for scikit-learn 1.5