compute_empirical_coverage#

bayesflow.utils.compute_empirical_coverage(estimates: ndarray, targets: ndarray, widths: ndarray, prob: float = 0.95, interval_type: str = 'central') dict[source]#

Compute empirical coverage statistics for given interval widths.

Parameters:
estimatesnp.ndarray of shape (num_datasets, num_post_draws, num_params)

The posterior draws obtained from num_datasets

targetsnp.ndarray of shape (num_datasets, num_params)

The true parameter values used for generating num_datasets

widthsnp.ndarray

Array of interval widths to compute coverage for (values between 0 and 1)

probfloat, optional, default: 0.95

Confidence level for coverage confidence intervals

interval_typestr, optional, default: “central”

Type of credible interval. Either “central” or “leftmost”

Returns:
dict

Dictionary containing coverage statistics for each width and parameter