Joint predictive distribution of a fitted model — predict

Draws the response for every row of newdata jointly: each simulation draws the coefficients from their (posterior) normal approximation, shared by all rows, then each row's response from the family. Rows are keyed row = 0, 1, .... An elastic_net_model has no standard errors, so its draws hold the coefficients fixed (process uncertainty only) and take a lambda from its path.

Usage

predict_distribution(object, newdata, n_sims, seed, ...)

Arguments

object: A glm_model, elastic_net_model, gam_model or bayes_glm_model (posterior predictive draws).
newdata: A data frame with the model's terms.
n_sims: Number of simulations.
seed: Generator seed.
...: Unused; for methods.
offset: Optional offset added to any offset() terms.
weights: Optional prior weights.
parameters: For a glm_model, how the coefficients are drawn: "normal" (beta ~ N(beta_hat, Sigma); through a log link the draws' mean is mu_hat * exp(x' Sigma x / 2)), "mean_preserving" (the same, with each row's linear predictor shifted so its draws average the fitted mean exactly; log or identity link) or "fixed" (process uncertainty only).

Value

A predictive_distribution.

Examples

d <- data.frame(y = c(2, 3, 5, 4, 6, 8), x = 1:6)
m <- glm_fit(y ~ x, d, family = "poisson")
pd <- predict_distribution(m, data.frame(x = c(7, 8)), n_sims = 1000, seed = 1)
mean(pd)
#> [1] 25.378