generate: Generate samples from fitted bru models

Description

Generic function for sampling for fitted models. The function invokes particular methods which depend on the class of the first argument.

Takes a fitted bru object produced by the function bru() and produces samples given a new set of values for the model covariates or the original values used for the model fit. The samples can be based on any R expression that is valid given these values/covariates and the joint posterior of the estimated random effects.

Usage

generate(object, ...)
# S3 method for bru
generate(
  object,
  newdata = NULL,
  formula = NULL,
  n.samples = 100,
  seed = 0L,
  num.threads = NULL,
  include = NULL,
  exclude = NULL,
  used = NULL,
  ...,
  data = deprecated()
)

Value

The form of the value returned by generate() depends on the data class and prediction formula. Normally, a data.frame is returned, or a list of data.frames (if the prediction formula generates a list)

List of generated samples

Arguments

object: A bru object obtained by calling bru().
...: additional, unused arguments.
newdata: A data.frame or SpatialPointsDataFrame of covariates needed for sampling.
formula: A formula where the right hand side defines an R expression to evaluate for each generated sample. If NULL, the latent and hyperparameter states are returned as named list elements. See Details for more information.
n.samples: Integer setting the number of samples to draw in order to calculate the posterior statistics. The default, 100, is rather low but provides a quick approximate result.
seed: Random number generator seed passed on to INLA::inla.posterior.sample
num.threads: Specification of desired number of threads for parallel computations. Default NULL, leaves it up to INLA. When seed != 0, overridden to "1:1"
include: Character vector of component labels that are needed by the predictor expression; Default: NULL (include all components that are not explicitly excluded) if newdata is provided, otherwise character(0).
exclude: Character vector of component labels that are not used by the predictor expression. The exclusion list is applied to the list as determined by the include parameter; Default: NULL (do not remove any components from the inclusion list)
used: Either NULL or a bru_used() object, overriding include and exclude.
data: Deprecated. Use newdata instead. sampling.

Details

In addition to the component names (that give the effect of each component evaluated for the input data), the suffix _latent variable name can be used to directly access the latent state for a component, and the suffix function _eval can be used to evaluate a component at other input values than the expressions defined in the component definition itself, e.g. field_eval(cbind(x, y)) for a component that was defined with field(coordinates, ...) (see also component_eval()).

For "iid" models with mapper = bru_mapper_index(n), rnorm() is used to generate new realisations for indices greater than n.

Examples

Run this code

# \donttest{
if (bru_safe_inla(multicore = FALSE) &&
    require("sn", quietly = TRUE)) {

  # Generate data for a simple linear model

  input.df <- data.frame(x = cos(1:10))
  input.df <- within(input.df, y <- 5 + 2 * cos(1:10) + rnorm(10, mean = 0, sd = 0.1))

  # Fit the model

  fit <- bru(y ~ xeff(main = x, model = "linear"),
    family = "gaussian", data = input.df
  )
  summary(fit)

  # Generate samples for some predefined x

  df <- data.frame(x = seq(-4, 4, by = 0.1))
  smp <- generate(fit, df, ~ xeff + Intercept, n.samples = 10)

  # Plot the resulting realizations

  plot(df$x, smp[, 1], type = "l")
  for (k in 2:ncol(smp)) points(df$x, smp[, k], type = "l")

  # We can also draw samples form the joint posterior

  df <- data.frame(x = 1)
  smp <- generate(fit, df, ~ data.frame(xeff, Intercept), n.samples = 10)
  smp[[1]]

  # ... and plot them
  if (require(ggplot2, quietly = TRUE)) {
    plot(do.call(rbind, smp))
  }
}
# }