non_credibility_index

probnumeval.multivariate.non_credibility_index(approximate_solution, reference_solution)

Compute the non-credibility index (NCI).

The NCI indicates how credible an estimate is. The smaller this value, the better. The NCI of a perfectly credible estimator is zero. Unlike the inclination index, the NCI cannot determine over- and underconfidence.

Parameters

• approximate_solution (_RandomVariableList) – Approximate solution as returned by a (Gaussian) probabilistic numerical method.

• reference_solution (ndarray) – Reference solution. This is an array, because it must be a deterministic point-estimate.

Returns

Return type

NCI statistic.

See also

anees()

An alternative calibration measure.

inclination_index()

A version of the NCI that can figure out over- or underconfidence.