Source code for skmetrics.regression

"""
Scikit-learn-compatible metrics for regression problems.

@author: David Diaz Vico
@license: MIT
"""

import numpy as np




def relative_mean_absolute_error(y_true, y_pred, sample_weight=None):
    """Relative mean absolute error

    Calculates the relative mean absolute error:
    100 * abs(y_true - y_pred) / abs(y_true).

    Parameters
    ----------
    y_true: array-like, shape = [n_samples]
            Ground truth (correct) target values.
    y_pred: array-like, shape = [n_samples]
            Estimated targets as returned by a classifier.
    sample_weight: array-like of shape = [n_samples], optional
                   Sample weights.

    Returns
    -------
    rmae: float
          Relative mean absolute error.
    """
    diff = np.abs(y_pred - y_true) / np.abs(y_true)
    rmae = np.average(diff, weights=sample_weight, axis=0)
    return rmae