rxmc.model_comparison.sharpness#

rxmc.model_comparison.sharpness(draws, percentiles=(16, 84), transform=None) → ndarray[source]#

Per-point width of a central predictive interval.

Parameters:

  • draws (array_like, shape (n_draws, n_pts))

  • percentiles ((float, float), optional) – Lower and upper percentiles (in 0-100) bounding the interval; the default is the central 68 %. Note coverage_curve() takes interval probabilities in (0, 1) instead.

  • transform (callable, optional) – Applied to the draws first (e.g. np.exp to report widths in raw space for a log comparison space, or np.log10).