spearman_correlation#

skfp.metrics.spearman_correlation(y_true: ndarray | list[float], y_pred: ndarray | list[float], *, alternative: str = 'two-sided', return_p_value: bool = False, equal_values_result: float = 1.0) → float#

Spearman correlation.

Calculates Spearman’s rank correlation coefficient (rho). It is a nonparametric measure of rank correlation. High value means that values of two variables change with monotonic relationship.

For constant inputs, i.e. exactly the same y_true and y_pred, 1.0 is returned. This differs from SciPy behavior, which returns NaN in that situation. This can be controlled with equal_values_result parameter.

Mainly intended for use as quality metric, where higher correlation between model prediction and ground truth is better. Can also be used for general correlation testing, by using alternative and return_p_value parameters.

Parameters:

  • y_true (array-like of shape (n_samples,) or (n_samples, n_outputs)) – Ground truth (correct) target values.

  • y_pred (array-like of shape (n_samples,) or (n_samples, n_outputs)) – Estimated target values.

  • alternative ({"two-sided", "less", "greater"}, default="two-sided") – Alternative hypothesis. By default, it checks if ground truth and estimated values differ in any direction.

  • return_p_value (bool, default=False) – Whether to return p-value instead of correlation value.

  • equal_values_result (float, default=1.0) – What value to return if y_true and y_pred are equal. This overrides the return_p_value parameter if necessary.

Returns:

score – Spearman correlation value, or p-value if return_p_value is True.

Return type:

float

Examples

>>> import numpy as np
>>> from skfp.metrics import spearman_correlation
>>> y_true = np.array([1, 2, 3])
>>> y_pred = np.array([3, 4, 5])
>>> spearman_correlation(y_true, y_pred)
1.0
>>> y_true = np.array([1, 2, 3, 4])
>>> y_pred = np.array([4, 3, 2, 1])
>>> spearman_correlation(y_true, y_pred)
-1.0