Module sample 

Observational and interventional batch sampling.

SPDX-License-Identifier: MIT OR Apache-2.0

Functions

refuse_cross_family_soft

Soft overrides must share noise semantics with the fitted mechanism. Reusing a Discrete Uniform(0,1) residual as an additive Gaussian U (or the reverse) is not a well-defined counterfactual.

sample_conditional_interventional

Sample under interventions conditioned on observed node values.

sample_interventional

Sample under interventions (compiled to an overlay; model is not cloned).

sample_observational

Sample n_rows observational draws from a fitted model.

sample_observational_into

Sample observational draws into a caller-owned column-major buffer.

sample_posterior_predictive

Posterior-predictive interventional sampling: for each coefficient draw block, refresh LinearGaussian slots then sample. draw_updater mutates slots in place.

sample_stochastic

Draw values from a stochastic intervention policy into out.

sample_structural_with_overlay

Structural path: sample noise then evaluate with overlays applied post-hoc for hard sets.

sample_with_overlay

Core ancestral sampler with overlay.

sample_with_overlay_into

Ancestral sample into a caller-owned column-major buffer.

soft_to_slot

Convert a soft MechanismOverride into a concrete mechanism slot.