Construct a visibility rule — visibility

A rule is a closure, not a table, and it is split into fit and predict. The split is what lets one interface cover exact derivation, donor means and (later) a fitted model, and what makes correct bootstrapping possible: a rule whose fit step consumes the sample must be refit inside each bootstrap replicate, and one whose does not must not be.

Usage

visibility_rule(
  label,
  requires,
  is_estimated,
  fit,
  predict,
  assumptions = character(0),
  params = list(),
  applies_to = NA_character_,
  tie_overridable = character(0),
  declared = list(),
  assumptions_fn = NULL
)

Arguments

label: short string naming the rule; appears in provenance output
requires: character vector of columns the rule needs, checked up front
is_estimated: TRUE if fit() consumes the sample. Drives the bootstrap path: see apply_visibility_rule()
fit: function(donor.dat, weights) returning opaque state
predict: function(alter.rows, state) returning one row per row of alter.rows, with columns vis, vis_weight and vis_rule
assumptions: character vector of assumptions this rule makes, carried into the provenance table so they reach output rather than living only in a methods appendix
params: list of the constructor's arguments, for printing

Value

an object of class visibility_rule