Nodes are the family's full sample space, weighted by the alternative's probabilities, so expectations are exact. Needs an enumerable sample space. Outcomes with zero probability are dropped.
Value
An engine spec, to pass as the engine argument of ripr_init().
Examples
fam <- multinomial_family(n_trials = 3L, k = 2L)
null <- null_model(fam, simplex_region(vertices = rbind(c(0.5, 0.5), c(0, 1))))
ripr_init(fam(c(0.6, 0.4)), null, engine = exact_engine())
#> <ripr_state>
#> null multinomial_family in 1 part
#> iterate 1 atom, KL 0.0604065
#> gap none recorded
#> steps fw 0, lb 0, em 0, weight 0, sfw 0, ct 0 (0.003s)
#> trace 1 row, 0 snapshots