R/utils.R

   1 # Helper for cross-validation: return the next test indices.
   2 get_testIndices <- function(n, CV, v, shuffle_inds) {
   3   if (CV$type == "vfold") {
   4     # Slice indices (optionnally shuffled)
   5     first_index = round((v-1) * n / CV$V) + 1
   6     last_index = round(v * n / CV$V)
   7     test_indices = first_index:last_index
   8     if (!is.null(shuffle_inds))
   9       test_indices <- shuffle_inds[test_indices]
  10   }
  11   else
  12     # Monte-Carlo cross-validation
  13     test_indices = sample(n, round(n * CV$test_size))
  14   test_indices
  15 }
  16
  17 # Helper which split data into training and testing parts.
  18 splitTrainTest <- function(data, target, testIdx) {
  19   dataTrain <- data[-testIdx,]
  20   targetTrain <- target[-testIdx]
  21   dataTest <- data[testIdx,]
  22   targetTest <- target[testIdx]
  23   # [HACK] R will cast 1-dim matrices into vectors:
  24   if (!is.matrix(dataTrain) && !is.data.frame(dataTrain))
  25     dataTrain <- as.matrix(dataTrain)
  26   if (!is.matrix(dataTest) && !is.data.frame(dataTest))
  27     dataTest <- as.matrix(dataTest)
  28   list(dataTrain=dataTrain, targetTrain=targetTrain,
  29        dataTest=dataTest, targetTest=targetTest)
  30 }