bqlearn.corruptions.uncertainty_noise_probability

bqlearn.corruptions.uncertainty_noise_probability(X, estimator, uncertainty='uncertainty', noise_ratio=0.5, random_state=None)

Get a probability of a sample to be noisy given an uncertainty function according to [1].

Valid uncertainty functions are:

[‘uncertainty’, ‘margin’, ‘entropy’, ‘density’]

Parameters:

Xarray-like of shape (n_samples, n_features)

The samples.

estimatorobject

The fitted estimator used to compute the noise probability. Must have attribute predict_proba or score_samples.

uncertainty{‘uncertainty’, ‘margin’, ‘entropy’, ‘density’}, default=’uncertainty’

Uncertainty function. ‘density’ is only available for estimators with score_samples attribute.

noise_ratiofloat, default=0.5

The ratio of noise. Must be between 0 and 1.

random_stateint or RandomState, default=None

Controls the noisy samples selection.

Returns:

noise_probabilitiesarray-like of shape (n_samples, )

The noise probabilities.

References

[1]

16. Nodet, V. Lemaire, A. Bondu, A. Cornuéjols, “Importance Reweighting for Biquality Learning”, IJCNN, 2021.