Predictive distribution from fitted means — simulate_from

Joint draws of the responses of n_rows rows, for engines that give a mean per row and nothing else (gradient boosting, an imported network): the family adds the process noise, and several mean vectors (bootstrap refits) add the parameter uncertainty. Simulation i picks one mean vector uniformly, then draws each row's response from family with that mean, dispersion and the row's weight. Components are keyed row = 0, 1, ..., as for glm_fit()'s draws.

Usage

simulate_from_means(
  family,
  means,
  n_sims,
  seed,
  n_rows = NULL,
  dispersion = 1,
  weights = NULL,
  theta = NULL,
  power = NULL
)

Arguments

family: "gaussian", "poisson", "gamma", "inverse_gaussian", "binomial", "negative_binomial" (needs theta) or "tweedie" (needs power).
means: A matrix with one row per policy and one column per mean vector, or a vector of n_rows * k means stacked column by column.
n_sims: Number of simulations.
seed: Generator seed.
n_rows: Rows per mean vector; defaults to nrow(means), or the length of a vector.
dispersion: The family's dispersion: one value, or one per row.
weights: Optional prior weights, one per row.
theta: Negative binomial theta.
power: Tweedie power in (1, 2).

Value

A predictive_distribution.

Examples

pd <- simulate_from_means("poisson", c(0.1, 0.4), n_sims = 20000, seed = 7)
mean(pd)
#> [1] 0.4955