Module sample 

Library-side orchestration for NUTS posterior sampling from a saved model.

The CLI’s gam sample subcommand and the Python Model.sample(...) API both call into sample_saved_model, which dispatches on the saved model’s class (standard GLM, standard with link-wiggle, or survival) and returns a fully-converged NutsResult over the original coefficient space. Gaussian identity standard models are sampled from the saved closed-form posterior, conditioning on the training fit rather than any prediction rows supplied by the caller.

Re-exports

pub use super::hmc_io::NutsConfig;

pub use super::hmc_io::NutsResult;

pub use super::hmc_io::PosteriorSampler;

Functions

laplace_gaussian_fallback

Draw iid samples from the fit’s PUBLISHED Gaussian posterior approximation N(mode, V), where V is the smoothing-corrected Vp whenever the fit carries one and the conditional Vb = cov_scale·H⁻¹ otherwise — the same choice summary() and the default predict(interval=...) make, so one fitted object publishes one posterior (gam#2777).

sample_saved_model

Run NUTS posterior sampling over a saved model.

saved_baseline_timewiggle_spec

Reconstruct the LinkWiggleFormulaSpec from a saved model’s baseline-time-wiggle runtime, returning None when the model has no time-wiggle component. Re-exported because the survival fitter’s tests exercise the spec independently of running NUTS.