InversePropensityWeighting.build

InversePropensityWeighting.build()

Construct the model graph without sampling anything.

Public, idempotent, and auto-called by both sampling verbs, so users never have to call it — but calling it explicitly makes the spec inspectable (pm.model_to_graphviz(exp.model), exp.model.basic_RVs, merged priors) before any compute is spent.

The inputs are recomputed from the experiment’s data on every call: for an already-built graph they are verified against the build-time fingerprint, so mutating exp.data and rebuilding raises RuntimeError instead of silently keeping a stale graph. The documented reset path is assigning a fresh model instance.

Returns:

The same experiment, for chaining.

Return type:

Self

Raises:

RuntimeError – If the graph exists and the experiment’s data changed since it was built.