pystra.active_learning.results.LearningStep#

Use it: User guide · Worked example · Theory

class LearningStep(failure_probability, learning_score, n_limit_state_evaluations, probability_band, beta_band, learning_satisfied, estimation_converged=True, sampling_cov=None, bootstrap_probability_band=None)[source]#

Bases: object

One fit’s probability diagnostics and cumulative true evaluation count.

probability_band is the ascending pair of probabilities for mean + 2*std <= 0 and mean - 2*std <= 0. The fixed MC workflow uses candidate proportions; adaptive estimators supply their own measure. beta_band transforms those endpoints into ascending reliability indices. These are surrogate sensitivity diagnostics, not confidence intervals or true Pf bounds. learning_satisfied is the selection policy’s threshold flag, before applying the stopping policy. The default requires an interior exploratory failure probability, under the estimator’s sampling measure. estimation_converged and optional sampling_cov record the exploratory estimator’s completion and precision, not the final sample. bootstrap_probability_band is the min/max of actual replicate Pf estimates on fixed IID normal enrichment, when requested. It need not enclose the full-design mean prediction and is not a confidence interval.

Methods

property beta: float#: Normal-equivalent index of the candidate probability.