R: Compute group.cv-values

inla.group.cv {INLA}

R Documentation

Compute group.cv-values

Description

From a fitted model, compute and add the group.cv-values

Usage

inla.group.cv(
  result,
  group.cv = NULL,
  num.level.sets = -1,
  strategy = c("posterior", "prior"),
  size.max = 32,
  groups = NULL,
  selection = NULL,
  group.selection = NULL,
  friends = NULL,
  weights = NULL,
  verbose = FALSE,
  epsilon = 0.005,
  prior.diagonal = 1e-04,
  keep = NULL,
  remove = NULL,
  remove.fixed = TRUE,
  type.cv = "single"
)

Arguments

result

An object of class inla, ie a result of a call to inla().

group.cv

If given, the groups are taken from this argument. group.cv must be the output of previous call to inla.group.cv().

num.level.sets

Number of level.sets to use. The default value -1 corresponds to leave-one-out cross-validation. If argument weights is used, then this is threshold for the sum of the weights defining a group.

strategy

One of "posterior" or "prior". See the vignette for details.

size.max

The maximum size (measure in the number of nodes) of a group. If the computed group-size is larger, it will be truncated to size.max. Note that: If weights are in use, then this still corresponds to the number of nodes in the group, and not the sum of the weights. This is ment as an emergency option to avoid the size of the group to go nuts.

groups

An (optional) predefined list of groups. See the vignette for details.

selection

An optional list of data-indices to use. If not given, then all data are used.

group.selection

An optional list of data-indices to use when building the groups. If given, each group beyond the observation itself, must be a subset of group.selection. If not given, then all data are used.

friends

An optional list of lists of indices to use a friends

weights

An optional positive weight attached to each datapoint. The sum of the weights define the size of the group. If NULL, then unit weight is used.

verbose

Run with verbose output of some of the internals in the calculations. This option will also enable inla(..., verbose=TRUE) if its not enabled already.

epsilon

Two correlations with a difference less than epsilon, will be classified as identical.

prior.diagonal

When strategy="prior", prior.diagonal is added to the diagonal of the prior precision matrix to avoid singularities

keep

For strategy="prior", then this gives a vector of the name of model-components TO USE when computing the groups. See the vignette for details. Not both of keep and remove can be defined.

remove

For strategy="prior", then this gives a vector of the name of model-components NOT TO USE when computing the groups. See the vignette for details. Not both of keep and remove can be defined.

remove.fixed

For strategy="prior", this is the default option which is in effect if both keep and remove are NULL. If TRUE, it will remove (or condition on) all fixed effects when computing the groups. See the vignette for details.

type.cv

Type of cv, either "single" (default) or "joint"

Value

The object returned is list related to leave-group-out cross-validation. See the vignette for details.

Author(s)