R/MAIL_Split_Stable.R

In hwyneken/MAILPackage: Model-Averaged Inferential Learning

#' MAIL_Split_Stable
#'
#' \code{MAIL_Split_Stable} runs MAIL with data splitting, and uses a more stable variable selection method
#'
#' This is a specific use of the \code{\link{MAIL}} function.
#' The function uses the following arguments with MAIL:
#' \itemize{
#'   \item splitOption = "Split"
#'   \item firstSOILWeightType = "BIC"
#'   \item smallestModelWeightType = "AIC"
#'   \item firstSOILPsi = 0.5
#'   \item smallestModelPsi = 0
#'   \item numSelectionIter = 100
#'   \item sigma2EstFunc = "Nested_LPM_AIC_CV_50Split"
#'   \item verbose = FALSE
#' }
#' @param XMat a n by p numeric matrix
#' @param yVec a n by 1 numeric vector
#' @return resList a list with the following elements:
#'
#'
#' \itemize{
#'   \item selectedSet: The indices of the selected variables (out of {1,...,p}).
#'   \item tempCI: A matrix (numSelected by 2) - each row corresponds to the 95% confidence interval for the full target of each selected variable (in the same order as selected set).
#'   \item margVar: A vector (length numSelected) of the sampling variances.
#'   \item betaHat: The vector (length p) of model-averaged estimates. Unselected variables have an estimated beta of zero.
#'   \item estSigma2: The estimated residual variance.
#'   \item candMat: An indicator matrix of candidate models (numModels by p).
#'   \item origSelectedSet: The variables selected the initial sweep by SOIL.
#' }
#'
#' @examples
#'
#' two main ideas to explain with an example
#' \itemize {
#'   \item how to run MAIL_Full
#'   \item how to run diagnostics for MAIL_Full - is it a reliable method for the current data
#' }
#'
#'
#' #### how to run MAIL_Split_Stable
#'
#' # first step - organize the data
#' # gene expression problem: response is the expression of gene probe X249.
#' # install package colonCA through Bioconductor (use these steps)
#' # https://bioconductor.org/packages/release/data/experiment/html/colonCA.html
#' require(colonCA)
#' data(colonCA)
#' dataSet <- t(colonCA@assayData$exprs)
#' colon_y <- dataSet[,249]
#' colon_x <- dataSet[,-1*249]
#' colon_x <- t(unique(t(colon_x))) # remove duplicate columns - original has p = 1999, reduce to p = 1990
#' colon_y <- scale(colon_y)
#' colon_x <- scale(colon_x)
#'
#' colonMAILFull <- MAIL_Split_Stable(XMat = colon_x,yVec = colon_y)
#'
#'
#' # compare the confidence intervals for MAIL_Full and MAIL_Split
#' # for X257
#'
#'
#' @export



#' @seealso \code{\link{MAIL}} and \code{\link{MAIL_Split}}

MAIL_Split_Stable = function(XMat,yVec) {
  
  
  resList = MAIL(XMat,yVec,
                 splitOption="Split",
                 firstSOILWeightType = "BIC",
                 smallestModelWeightType = "AIC",
                 firstSOILPsi = 0.5,
                 smallestModelPsi = 0,
                 numSelectionIter = 100,
                 sigma2EstFunc = "Nested_LPM_AIC_CV_50Split",
                 trueSD = NULL,
                 verbose=FALSE)
  ##
  rownames(resList$tempCI) <- colnames(XMat)[resList$selectedSet]
  colnames(resList$tempCI) <- c("95% CI: Lower Bound",
                                "95% CI: Upper Bound")
  
  return(resList)
}