pystra.active_learning.importance.ImportanceSamplingDiagnostics

Use it: User guide · Worked example · Theory

class ImportanceSamplingDiagnostics(raw_probability, standard_error, effective_samples, effective_failures, failure_count, mean_weight, max_weight)

Bases: object

IID likelihood-weight diagnostics conditional on a frozen surrogate.

effective_failures uses failure contributions w*I, while effective_samples uses all weights. Neither detects unsampled modes. raw_probability retains an out-of-range estimate if a finite sample cannot give a valid probability; such results are explicitly incomplete.

Methods