pystra.active_learning.learning.FBRLearning

Use it: User guide · Worked example · Theory

class FBRLearning

Bases: EnsembleLearningFunction

Minimum bootstrap classification agreement, abs(B_safe-B_failure)/B.

Marelli & Sudret (2018), Eq. (10), doi:10.1016/j.strusafe.2018.06.003. Zero is maximal disagreement; one is unanimity. Zero response counts as failure. Ties select the first candidate. The threshold flag means all candidates are unanimous; it is not a guarantee of surrogate accuracy. Use BootstrapBounds for the paper’s probability-range stopping rule.

Methods

select	Reject lossy mean/std input; use select_replicates instead.
select_replicates	Use actual classification votes, retaining non-Gaussian behavior.

select_replicates(predictions)

Use actual classification votes, retaining non-Gaussian behavior.

Return type:

LearningDecision

select(mean, std)

Reject lossy mean/std input; use select_replicates instead.

Return type:

LearningDecision