pystra.active_learning

Use it: User guide · Worked example · Theory

Composable active-learning methods for structural reliability.

The four components are a surrogate, reliability estimator, learning function and stopping criterion; see Moustapha, Marelli and Sudret (2022), doi:10.1016/j.strusafe.2021.102174, and Teixeira, Nogal and O’Connor (2021), doi:10.1016/j.strusafe.2020.102019. Enrichment uses a fixed normal MC pool by default; SubsetSimulationEstimator generates adaptive conditional populations with its own probability diagnostics and independent final estimation. Kriging requires the optional al extra. PCE uses only NumPy/SciPy.

analysis	Compose active-learning reliability components.
estimation	Final reliability estimation conditional on a frozen surrogate.
importance	Active importance sampling with explicit Gaussian mixture proposals.
learning	Point selection in independent standard normal coordinates.
pc_kriging	Sequential PC-Kriging with a selected Hermite trend and universal variance.
results	Immutable records separating surrogate diagnostics and sampling error.
stopping	Stateless stopping policies evaluated against immutable fit histories.
subset	Replicated subset simulation and estimator-driven active learning.
surrogates	Optional active-learning reliability in independent normal coordinates.