diff --git a/data_preprocessing/k_folds.r b/data_preprocessing/k_folds.r index 37ea54a8..12fb62a2 100644 --- a/data_preprocessing/k_folds.r +++ b/data_preprocessing/k_folds.r @@ -1,29 +1,49 @@ -# K folds cross validation is essential for machine learning -# createFolds function in package caret is easy to use -# here we write our own function +# Create randomized k-fold cross-validation assignments. +# +# y: vector whose observations should be divided into folds +# k: number of folds +# isList: if TRUE, return observation indices grouped by fold; +# if FALSE, return one fold ID per observation +# seed: random seed for reproducibility -get_k_folds<-function(y = c(),k = 10, isList = TRUE, seed = 123){ +get_k_folds <- function( + y = c(), + k = 10, + isList = TRUE, + seed = 123 +) { + n <- length(y) + + if (n == 0) { + stop("y must contain at least one observation") + } + + if (length(k) != 1 || + !is.numeric(k) || + is.na(k) || + k != as.integer(k)) { + stop("k must be a positive integer") + } + + if (k < 2 || k > n) { + stop("k must be between 2 and length(y)") + } + + if (length(isList) != 1 || !is.logical(isList) || is.na(isList)) { + stop("isList must be TRUE or FALSE") + } + set.seed(seed) - folds<-sample(1:length(y), length(y)) - every_n<-ceiling(length(y)/k) - matFolds<-suppressWarnings(matrix(folds, ncol=every_n, byrow = T)) - - if(isList){ - value<-NULL - rownames(matFolds)<-paste("Folds",1:k,sep="") - value<-lapply(1:k, function(x){ - if(x == k){ - return(matFolds[x,][1:(length(y)-every_n*(k-1))]) - }else{ - return(matFolds[x,]) - } - }) - }else{ - value<-c() - for(i in 1:length(y)){ - value[i]<-ceiling(i/every_n) - } + + # Randomly assign every observation to a fold. + # Fold sizes differ by at most one observation. + fold_ids <- sample(rep(seq_len(k), length.out = n)) + + if (isList) { + return(lapply(seq_len(k), function(fold) { + which(fold_ids == fold) + })) } - return(value) -} + fold_ids +} \ No newline at end of file