bru_set_missing: Set missing values in observation models

In inlabru-org/inlabru: Bayesian Latent Gaussian Modelling using INLA and Extensions

Set missing values in observation models

Description

Set all or parts of the observation model response data to NA, for example for use in cross validation (with bru_rerun()) or prior sampling (with bru_rerun() and generate(), but see "Prior sampling caveats" below).

Usage

bru_set_missing(object, keep = FALSE, ...)

## S3 method for class 'bru'
bru_set_missing(object, keep = FALSE, ...)

## S3 method for class 'bru_obs_list'
bru_set_missing(object, keep = FALSE, ...)

## S3 method for class 'bru_obs'
bru_set_missing(object, keep = FALSE, ...)

## Default S3 method:
bru_set_missing(object, keep = FALSE, ...)

## S3 method for class 'data.frame'
bru_set_missing(object, keep = FALSE, ...)

## S3 method for class 'inla.surv'
bru_set_missing(object, keep = FALSE, ...)

Arguments

object	A bru, bru_obs or bru_obs_list object
keep	For bru_obs, a single logical or an integer vector; If TRUE, keep all the response data, if FALSE (default), set all of it to NA. An integer vector determines which elements to keep (for positive values) or to set as missing (negative values). For bru and bru_obs_list, a logical scalar or vector, or a list, see Details.
...	Additional arguments passed on to the bru_obs and individual data class methods. Currently unused.

Details

For bru and bru_obs_list,

• keep must be either a single logical, which is expanded to a list,

• a logical vector, which is converted to a list,

• an unnamed list of the same length as the number of observation models, with elements compatible with the bru_obs method, or

• a named list with elements compatible with the bru_obs method, and only the named bro_obs models are acted upon, i.e. the elements not present in the list are treated as keep = TRUE.

E.g.: keep = list(b = FALSE) sets all observations in model b to missing, and does not change model a.

E.g.: keep = list(a = 1:4, b = -(3:5)) keeps only observations 1:4 of model a, marking the rest as missing, and sets observations 3:5 of model b to missing.

Methods (by class)

• bru_set_missing(default): From ⁠> 2.13.0⁠, set missing values in any object supporting base::is.na<-() for positive and negative indices.

• bru_set_missing(data.frame): From ⁠> 2.13.0⁠, handles data.frame, tibbles, including inla.mdata.

• bru_set_missing(inla.surv): From ⁠> 2.13.0⁠, handles inla.surv.

Prior sampling caveats

Note that prior sampling requires special care for hyperparameters, as the prior modes are not typically useful; in the future, we plan to have a dedicated method that samples from the hyperparameters, and then uses

bru_rerun(
  bru_set_missing(...),
  options = list(
    control.mode = list(
      theta = theta_sample,
      fixed = TRUE)
    )
  )

for each sample.

Examples

obs <- c(
  A = bru_obs(y_A ~ ., data = data.frame(y_A = 1:6)),
  B = bru_obs(y_B ~ ., data = data.frame(y_B = 11:15))
)
bru_response_size(obs)
lapply(
  bru_set_missing(obs, keep = FALSE),
  function(x) {
    x[["response_data"]][["BRU_response"]]
  }
)
lapply(
  bru_set_missing(obs, keep = list(B = FALSE)),
  function(x) {
    x[["response_data"]][["BRU_response"]]
  }
)
lapply(
  bru_set_missing(obs, keep = list(1:4, -(3:5))),
  function(x) {
    x[["response_data"]][["BRU_response"]]
  }
)


(obs <- INLA::inla.mdata(y = 1:4, X = matrix(1:8, 4, 2)))
bru_set_missing(obs, keep = c(1, 4))


(obs <- INLA::inla.surv(time = 1:4, event = c(1, 0, 1, 0)))
bru_set_missing(obs, keep = c(1, 4))

(obs <- INLA::inla.surv(
  time = 1:4,
  event = c(1, 0, 1, 0),
  cure = matrix(1:8, 4, 2)
))
bru_set_missing(obs, keep = c(1, 4))|>str()

(obs <- INLA::inla.surv(
  time = 1:4,
  event = c(1, 0, 1, 0),
  subject = c(1, 1, 2, 1)
))
bru_set_missing(obs, keep = c(1, 4))

inlabru-org/inlabru documentation built on July 17, 2025, 2:11 a.m.