Models and analysis options#

Define random variables, limit states, model constants and algorithm options.

Public entry points#

Object	Purpose
`StochasticModel`	Collect named variables, constants and dependence.
`LimitState`	Wrap the physical response and gradient contract.
`pystra.options`	Frozen FORM, SORM and simulation settings.

Use it: Specifying a reliability model · A first reliability analysis · Structural reliability fundamentals

Module details#

`pystra.model`	Named stochastic models and physical limit-state evaluation.
`pystra.reliability.analysis`	Base class for reliability analyses.
`pystra.options`	Immutable settings for reliability analyses.
`pystra.errors`	Exceptions raised by PySTRA.

Limit-state evaluation#

LimitState.evaluate(x, model, differentiation="ffd") accepts a physical point (n_variables,) or rows of points (n_samples, n_variables). Values are scalar for a point and (n_samples,) for a batch; gradients have the input shape. Columns and gradient entries follow model random-variable order, excluding constants. Square batches always contain rows of points. differentiation="ddm" uses the expression’s analytic gradient; "no" returns zero gradients. Failure is g <= 0.

User-script migration#

python -m pystra.migrate path/to/script.py path/to/notebook.ipynb previews unified diffs for 1.x code. --write applies the same changes. Directories are scanned for Python files and notebooks. The converter preserves local identifiers, string literals and comments, and changes only resolved PySTRA imports, module attributes and constructor keywords. Shadowed names remain unchanged. Instance methods, result getters, options, parameter modes and calibration workflows require manual review; diagnostics identify their locations. A second pass must leave the source unchanged.

Notebook imports carry across valid Python cells in document order. Cells containing IPython syntax are left unchanged and flagged. Only changed code source fields are serialized; Markdown, outputs and metadata are preserved. Exit status is zero with no review items, one when manual review is needed, and two for file or input errors. A successful conversion is not a guarantee that the resulting analysis is runnable or numerically appropriate.