ThresholdAdjustment

class mlquantify.counting.ThresholdAdjustment(estimator=None, threshold=0.5, strategy='ovr', n_jobs=None, cv=5, stratified=True, shuffle=False, random_state=None)

Abstract base class for ROC-threshold adjustment quantifiers.

Corrects the bias in CC estimates by selecting a threshold on the ROC curve and adjusting the observed positive proportion using the corresponding TPR and FPR. Subclasses implement get_best_threshold to define the selection strategy.

This is a binary-only method. When applied to multiclass problems, a one-vs-rest (OvR) strategy is applied automatically.

Parameters:

estimatorestimator, optional

A classifier with fit and predict_proba methods.

thresholdfloat, default=0.5

Default classification threshold.

strategy{‘ovr’, ‘ovo’}, default=’ovr’

Multiclass decomposition strategy.

n_jobsint or None, default=None

Number of parallel jobs for multiclass decomposition.

Attributes:

estimator_estimator

The fitted underlying classifier.

classes_ndarray of shape (n_classes,)

Class labels seen during fit.

train_predictionsndarray

Cross-validated soft predictions used for TPR/FPR estimation.

y_trainndarray of shape (n_samples,)

Training labels corresponding to train_predictions.

References

References

[1]

Forman, G. (2005). Counting Positives Accurately Despite Inaccurate Classification. ECML, pp. 564–575.

[2]

Forman, G. (2008). Quantifying Counts and Costs via Classification. Data Mining and Knowledge Discovery, 17(2), 164–206.

Examples

>>> from mlquantify.counting import ThresholdAdjustment
>>> from sklearn.ensemble import RandomForestClassifier
>>> from sklearn.datasets import make_classification
>>> class CustomTA(ThresholdAdjustment):
...     def get_best_threshold(self, thresholds, tprs, fprs):
...         idx = np.argmin(np.abs(tprs - 0.5))
...         return thresholds[idx], tprs[idx], fprs[idx]
>>> X, y = make_classification(n_samples=100, n_classes=2, n_informative=5, random_state=42)
>>> quantifier = CustomTA(estimator=RandomForestClassifier(random_state=42))
>>> quantifier.fit(X, y)
>>> quantifier.predict(X)
array([[0.3, 0.7]])

aggregate(predictions, y_train)

Aggregate posteriors into prevalences using MoSS score simulation.

Searches over merging_factors to find the synthetic score distribution (generated by MoSS) whose histogram is closest to the test score distribution, then passes that synthetic set as the training reference to the base quantifier’s aggregate.

Parameters:

predictionsndarray of shape (n_samples, n_classes)

Posterior probabilities of the test instances.

y_trainndarray of shape (n_train_samples,)

Training class labels used to resolve class ordering.

Returns:

prevalencesdict or ndarray of shape (n_classes,)

Estimated class prevalences.

Examples

>>> from mlquantify.meta import QuaDapt
>>> from mlquantify.matching import DyS
>>> from sklearn.linear_model import LogisticRegression
>>> from sklearn.datasets import make_classification
>>> X, y = make_classification(n_samples=200, random_state=42)
>>> q = QuaDapt(DyS(LogisticRegression())).fit(X, y)
>>> proba = LogisticRegression().fit(X, y).predict_proba(X)
>>> q.aggregate(proba, y)
{0: 0.49, 1: 0.51}

fit(X, y)

Fit the base classifier of the wrapped quantifier.

Only the underlying estimator is trained here; the full aggregation is deferred to aggregate so that the MoSS-based correction can be applied at prediction time.

Parameters:

Xarray-like of shape (n_samples, n_features)

Training feature matrix.

yarray-like of shape (n_samples,)

Training class labels.

Returns:

selfQuaDapt

The fitted quantifier.

Raises:

ValueError

If the wrapped quantifier does not use soft (probabilistic) predictions.

Examples

>>> from mlquantify.meta import QuaDapt
>>> from mlquantify.matching import DyS
>>> from sklearn.linear_model import LogisticRegression
>>> from sklearn.datasets import make_classification
>>> X, y = make_classification(n_samples=200, random_state=42)
>>> q = QuaDapt(DyS(LogisticRegression())).fit(X, y)

fit_predict(X, y, X_test)

Fit and predict class prevalences without storing models.

abstract get_best_threshold(thresholds, tprs, fprs)

Select the best threshold according to the specific method.

get_metadata_routing()

Get metadata routing of this object.

Please check User Guide on how the routing mechanism works.

Returns:

routingMetadataRequest

A MetadataRequest encapsulating routing information.

get_params(deep=True)

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)

Predict class prevalences using the MoSS adaptive correction.

Generates posterior probabilities for X with the fitted classifier and delegates to aggregate, which selects the best MoSS merging factor and calls the base quantifier’s aggregate.

Parameters:

Xarray-like of shape (n_samples, n_features)

Test feature matrix.

Returns:

prevalencesdict or ndarray of shape (n_classes,)

Estimated class prevalences.

Examples

>>> from mlquantify.meta import QuaDapt
>>> from mlquantify.matching import DyS
>>> from sklearn.linear_model import LogisticRegression
>>> from sklearn.datasets import make_classification
>>> X, y = make_classification(n_samples=200, random_state=42)
>>> q = QuaDapt(DyS(LogisticRegression())).fit(X, y)
>>> q.predict(X)
{0: 0.49, 1: 0.51}

save_quantifier(path: str | None = None) → None

Save the quantifier instance to a file.

set_fit_request(*, cv_prediction: bool | None | str = '$UNCHANGED$', estimator_fitted: bool | None | str = '$UNCHANGED$') → ThresholdAdjustment

Configure whether metadata should be requested to be passed to the fit 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 fit 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 fit.

• 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:

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

Metadata routing for cv_prediction parameter in fit.

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

Metadata routing for estimator_fitted parameter in fit.

Returns:

selfobject

The updated object.

set_params(**params)

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.