pystra.active_learning.estimation.ReliabilityEstimator

Use it: User guide · Worked example · Theory

class ReliabilityEstimator

Bases: ABC

Stateless final estimator using independent standard normal coordinates.

The supplied predictor accepts row-wise (n_samples, n_variables) points and returns mean/std arrays. Failure is mean <= 0. It does not evaluate the true model. Use only the run-owned rng for randomness, and keep samples separate from training/enrichment. Estimators own their sample weights and dependence-aware uncertainty calculations: the runner never replaces those diagnostics with a binomial formula.

This base interface controls the final estimate, retaining fixed normal MC enrichment. Subclass EnrichmentEstimator to also generate enrichment populations and probability diagnostics under the estimator’s measure.

Methods

estimate

Estimate conditional Pf from a frozen surrogate, returning a snapshot.

abstractmethod estimate(predict, *, dimension, rng)

Estimate conditional Pf from a frozen surrogate, returning a snapshot.

Return type:

ReliabilityEstimate