cgeneric_graphpcor: Build an 'inla.cgeneric' for a graph, see 'graphpcor()'
In graphpcor: Models for Correlation Matrices Based on Graphs

cgeneric_graphpcor

R Documentation

Build an `inla.cgeneric` for a graph, see `graphpcor()`

Description

From either a graph (see graph()) or a square matrix (used as a graph), creates an inla.cgeneric (see cgeneric()) to implement the Penalized Complexity prior using the Kullback-Leibler divergence - KLD from a base graphpcor.

Usage

cgeneric_graphpcor(
  graph,
  lambda,
  base,
  sigma.prior.reference,
  sigma.prior.probability,
  params.id,
  low.params.fixed,
  debug = FALSE,
  useINLAprecomp = TRUE,
  libpath = NULL
)

Arguments

`graph`	a `graphpcor` (see `graphpcor()`) or a square matrix (to be used as a graph) to define the precision structure of the model.
`lambda`	the parameter for the exponential prior on the radius of the sphere, see details.
`base`	numeric vector with length `m`, `m` is the number of edges in the graph, or matrix with the reference correlation model against what the KLD will be evaluated. If it is a vector, a correlation matrix is defined considering the graph model and this vector as the parameters in the lower triangle matrix L. If it is a matrix, it will be checked if the graph model can generates this.
`sigma.prior.reference`	numeric vector with length `n`, `n` is the number of nodes (variables) in the graph, as the reference standard deviation to define the PC prior for each marginal variance parameters. If missing, the model will be assumed for a correlation. If a length `n` vector is given and `sigma.prior.reference` is missing, it will be used as known square root of the variances. NOTE: `params.id` will be applied here as `sigma.prior.reference[params.id[1:n]]`.
`sigma.prior.probability`	numeric vector with length `n` to set the probability statement of the PC prior for each marginal variance parameters. The probability statement is P(sigma < `sigma.prior.reference`) = p. If missing, all the marginal variances are considered as known, as described in `sigma.prior.reference`. If a vector is given and a probability is NA, 0 or 1, the corresponding `sigma.prior.reference` will be used as fixed. NOTE: `params.id` will be applied here as `sigma.prior.probability[params.id[1:n]]`.
`params.id`	integer ordered vector with length equals to `n+m` to specify common parameter values. If missing it is assumed `1:(n+m)` and all parameters are assumed distinct. The first `n` indexes the square root of the marginal variances and the remaining indexes the edges parameters. Example: By setting `params.id = c(1,1,2,3, 4,5,5,6)`, the first two standard deviations are common and the second and third edges parameters are common as well, giving 6 unknown parameters in the model.
`low.params.fixed`	numeric vector of length `m` providing the value(s) at which the lower parameter(s) of the L matrix to be fixed and not estimated. NA indicates not fixed and all are set to be estimated by default. Example: with `low.params.fixed = c(NA, -1, NA, 1)` the first and the third of these parameters will be estimated while the second is fixed and equal to -1 and the forth is fixed and equal to 1. NOTE: `params.id` will be applied here as `low.params.fixed[params.id[(n+1:m)]-n+1]`, thus the provided examples give `⁠NA -1 -1 NA⁠` and so the second and third low L parameters are fixed to `-1`.
`debug`	integer, default is zero, indicating the verbose level. Will be used as logical by INLA.
`useINLAprecomp`	logical, default is TRUE, indicating if it is to be used the shared object pre-compiled by INLA. This is not considered if 'libpath' is provided.
`libpath`	string, default is NULL, with the path to the shared object.