auc_mu

lightautoml.tasks.common_metric.auc_mu(y_true, y_pred, sample_weight=None, class_weights=None)[source]

Compute multi-class metric AUC-Mu.

We assume that confusion matrix full of ones, except diagonal elements. All diagonal elements are zeroes. By default, for averaging between classes scores we use simple mean.

Parameters:

  • y_true (ndarray) – True target values.

  • y_pred (ndarray) – Predicted target values.

  • sample_weight (Optional[ndarray]) – Not used.

  • class_weights (Optional[ndarray]) – The between classes weight matrix. If None, the standard mean will be used. It is expected to be a lower triangular matrix (diagonal is also full of zeroes). In position (i, j), i > j, there is a partial positive score between i-th and j-th classes. All elements must sum up to 1.

Return type:

float

Returns:

Metric value.

Note

Code was refactored from https://github.com/kleimanr/auc_mu/blob/master/auc_mu.py