R/splitCommon.R

In pcahan1/cancerCellNet: CellNet, for cancer

#' @title
#' Split Sample Table
#'
#' @description
#' Split the sample table into training and validation through random sampling.
#'
#' @param sampTab sample table
#' @param ncells number of samples for training in each category. If left empty, will automatically select half of the number of samples in the smallest category.
#' @param dLevel the column name with the classification categories.
#' @return a list containing training sample table and validation sample table
#'
#' @export
splitCommon <- function(sampTab, ncells=NA, dLevel) {
  cts = unique(as.vector(sampTab[,dLevel])) #receive the names of the categories
  trainingids = vector()

  # automatically pick smallest number of samples in category / 2  if ncell value is missing
  if(is.na(ncells) == TRUE) {
    smallestNCells = 1e9

    for(ct in cts) {
      if(nrow(sampTab[sampTab[,dLevel]==ct,]) < smallestNCells) {
        smallestNCells = nrow(sampTab[sampTab[,dLevel]==ct,])
      }
    }

    ncells = as.integer(smallestNCells / 2) # select the ncell based on the smallest samples divided by 2
  }

  for(ct in cts) {
    stX = sampTab[sampTab[,dLevel]==ct,]

    # Error catching mechanism if the ncells exceeds the smallest sample size
    if(nrow(stX) < ncells) {
      stop(paste0("Category ", ct, " has ", nrow(stX), " samples. Please pick a samller number for ncells. "))
    }
    if(ncells <= 0) {
      stop(paste0("Category ", ct, " has ", nrow(stX), " samples. Please pick a number above 0 for ncells. "))
    }
    cat(ct, "has ", nrow(stX), " samples.", "\n")

    trainingids = append(trainingids, sample(rownames(stX), ncells)) # randomly samples ccount of training samples
  }

  val_ids = setdiff(rownames(sampTab), trainingids) # the samples that are not used to training are used for validation

  return(list(train=sampTab[trainingids,], val=sampTab[val_ids,]))
}