R/BicZ.R

In CorReg: Linear Regression Based on Linear Structure Between Variables

#' Compute the BIC of a given structure
#' 
#' @param X the dataset
#' @param Z binary adjacency matrix of the structure (size p)
#' @param Bic_null_vect the BIC of the null hypothesis (used for independent variables)
#' @param Bic_old BIC (vector) associated to Zold
#' @param Zold another structure with some common parts with Z (allows to compute only the differences, to be faster)
#' @param methode parameter for OLS (matrix inversion) methode_BIC  parameter for OLS (matrix inversion) 1:householderQr, 2:colPivHouseholderQr
#' @param star boolean defining wether classical BIC or BIC* (over-penalized by a hierarchical uniform assumption to avoid over-learning)is computed
#' @return The vector of the BICs associated to each covariate (conditionnal distribution) according to the sub-regression structure.
#' 
#' @examples
#' data = mixture_generator(n = 15, p = 5, valid = 0) # dataset generation
#' Z = data$Z # binary adjacency matrix that describes correlations within the dataset
#' X = data$X_appr
#' Bic_null_vect = density_estimation(X = X)$BIC_vect
#' # Computes the BIC associated to each covariate (optional, BicZ can do it if not given as an input)
#' # computes the BIC associated to the structure
#' res = BicZ(X = X, Z = Z, Bic_null_vect = Bic_null_vect)
#'
#' @export
BicZ <- function(X = X, Z = Z, Bic_null_vect = NULL, Bic_old = NULL, methode = 1, Zold = NULL, star = FALSE) {
   if (is.null(Bic_null_vect)) {
      Bic_null_vect = density_estimation(X = X)$BIC_vect
   }
   if (is.null(Zold) | is.null(Bic_old)) {
      Zold = 0 * Z
      Bic_old = Bic_null_vect
   }
   res = .Call("BicZ", as.matrix(X), Z, Bic_null_vect, Bic_old, methode, Zold, PACKAGE = "CorReg")
   if (star) {
      res$BIC = sum(res$BIC) - ProbaZ(Z, star = TRUE)
   }
   
   return(res$BIC)
}