R/MAVE_CV.R

In MAVE: Methods for Dimension Reduction

#' Select best direction using cross-validation
#'
#' This function selects the dimension of the central (mean)  space
#' based on the calculation of MAVE  using cross-validation method.
#'
#' @param dr the result of MAVE function
#' @param max.dim the maximum dimension for cross-validation.
#'
#' @return dr.dim contains all information in dr plus cross-validation values of corresponding
#' direction
#' \itemize{
#' \item cv0 : the cross-validation value when the null model is used
#' \item cv : the cross-validation value using dimension reduction directions of different dimensions
#' \item dim.min : the dimension of minimum cross-validation value. Note that this value can be 0.
#' }
#' @seealso \code{\link{mave}} for computing the dimension reduction space, \code{\link{predict.mave.dim}}
#' for prediction method of mave.dim class
#' @export
#'
#' @examples
#'  x <- matrix(rnorm(400*5),400,5)
#'  b1 <- matrix(c(1,1,0,0,0),5,1)
#'  b2 <- matrix(c(0,0,1,1,0),5,1)
#'  eps <- matrix(rnorm(400),400,1)
#'  y <- x%*%b1 + (x%*%b2)*eps
#'
#'  #seleted dimension of central space
#'  dr.cs <- mave(y~x,method='csmave')
#'  dr.cs.dim <- mave.dim(dr.cs)
#'
#'  #seleted dimension of central mean space
#'  dr.mean <- mave(y~x,method='meanmave')
#'  dr.mean.dim <- mave.dim(dr.mean)

mave.dim<-function(dr, max.dim=10){

  max.dim = min(max.dim, ncol(dr$x))
  which.dim <- 1:min(dr$max.dim, max.dim)
  if(length(which.dim)==0){
    warning('No CV is computed given the dr$max.dim and the argument max.dim')
    return(dr)
  }
  if(is.null(dr$cv)){
    cv <- rep(Inf, max.dim)
  }
  
  n = nrow(dr$ky)
  cv0 = mean(abs(dr$ky)*(1+1/n))
  
  for(dim in which.dim){
    cv[dim] <- CVfastCpp(mave.data(dr,dr$x,dim),dr$ky)
  }
  dr$cv = cv
  dr$cv0 = cv0
  dr$dim.min = which.min(c(cv0,cv))-1
  dr$call<-match.call()
  class(dr)<-'mave.dim'
  return(dr)
}