R/splitFolds.R

In jchitpin/blistR:

splitFolds <- function(indexFeature, DT_folds, feat, folds){

    ## Define training set output
    DT_training <- vector("list", length=7)

    ## Populate testing dataset based on 'folds'
    DT_testing <- DT_folds[[feat]][[folds]]

    ## Populate tuning dataset based on 'folds'.
    ## Visually, this is the next set after the testing dataset
    ## 'toCombine' is an index to populate training dataset based on NOT 'folds'
    if(folds < length(DT_folds[[feat]])){
        DT_tuning <- DT_folds[[feat]][[folds+1]]
        toCombine <- which(!(seq_along(DT_folds[[feat]]) %in% c(folds, folds+1)))
    } else{
        DT_tuning <- DT_folds[[feat]][[1]]
        toCombine <- which(!(seq_along(DT_folds[[feat]]) %in% c(1, folds)))
    }

    ## Push remaining folds into a single training dataset
    for(combining in seq_along(toCombine)){
        ## First iteration, initialize 'DT_training' to first fold
        if(combining == 1){
            DT_training[[feat]] <- DT_folds[[feat]][[toCombine[combining]]]
        } else{
            ## Strip non-feature-related columns in subsequent folds and column bind
            minimumCol <- grep("ExactMass", colnames(DT_folds[[feat]][[toCombine[combining]]]))
            temp <- DT_folds[[feat]][[toCombine[combining]]]
            temp <- temp[, (minimumCol+1):ncol(temp)]
            DT_training[[feat]] <- cbind(DT_training[[feat]], temp)
        }
    }

    return(list(testing  = DT_testing,
                tuning   = DT_tuning,
                training = DT_training[[feat]]))
}