Methods to score the quality of each label in a regression dataset. These can be used to rank the examples whose Y-value is most likely erroneous.

Note: Label quality scores are most accurate when they are computed based on out-of-sample predictions from your regression model. To obtain out-of-sample predictions for every datapoint in your dataset, you can use cross-validation. This is encouraged to get better results.

If you have a sklearn-compatible regression model, consider using cleanlab.regression.learn.CleanLearning instead, which can more accurately identify noisy label values.


get_label_quality_scores(labels, predictions, *)

Returns label quality score for each example in the regression dataset.

cleanlab.regression.rank.get_label_quality_scores(labels, predictions, *, method='outre')[source]#

Returns label quality score for each example in the regression dataset.

Each score is a continous value in the range [0,1]

  • 1 - clean label (given label is likely correct).

  • 0 - dirty label (given label is likely incorrect).

  • labels (array_like) – Raw labels from original dataset. 1D array of shape (N, ) containing the given labels for each example (aka. Y-value, response/target/dependent variable), where N is number of examples in the dataset.

  • predictions (np.ndarray) – 1D array of shape (N,) containing the predicted label for each example in the dataset. These should be out-of-sample predictions from a trained regression model, which you can obtain for every example in your dataset via cross-validation.

  • method ({"residual", "outre"}, default "outre") – String specifying which method to use for scoring the quality of each label and identifying which labels appear most noisy.

Return type:



label_quality_scores – Array of shape (N, ) of scores between 0 and 1, one per example in the dataset.

Lower scores indicate examples more likely to contain a label issue.


>>> import numpy as np
>>> from cleanlab.regression.rank import get_label_quality_scores
>>> labels = np.array([1,2,3,4])
>>> predictions = np.array([2,2,5,4.1])
>>> label_quality_scores = get_label_quality_scores(labels, predictions)
>>> label_quality_scores
array([0.00323821, 0.33692597, 0.00191686, 0.33692597])