Select sample from train data based on fold from k-fold cross-validation

Question

I have performed the k-fold cross-validation without package based on here How to split a data set to do 10-fold cross validation using no packages

I need to select 30% of the sample from each fold in train data. Here is my function:

samples = 300
r = 0.83

library('MASS')

df = data.frame(mvrnorm(n=samples, mu=c(0, 0), Sigma=matrix(c(1, r, r, 1), nrow=2), empirical=TRUE))
w = df[sample(nrow(df)),]
w = data.frame(w)
kcv = 10
folds <- cut(seq(from = 1,to = nrow(w)),breaks=kcv,labels=FALSE)
kfolddata<-cbind(w,folds)


for(i in 1:kcv){ #i=1
  testIndexes <- which(kfolddata[,ncol(kfolddata)]==i,arr.ind=TRUE)
  testData <- w[testIndexes, ]
  trainData <- w[-testIndexes, ]
  trainIndexes <- kfolddata[-testIndexes,]

  if(i==1) {
    set.seed=1234
    SubInd = sample(nrow(trainData) , size = round(0.3 * 
                                (nrow(trainData))),replace=F)
  } else {
     SubInd = rbind(SubInd,sample(nrow(trainData) , size = round(0.3 *
                                 nrow(trainData))),replace=F))}}
  }
}

The results will only display the ID of the selected subset. How can I obtain the information (the variables) for the selected ID for SubInt?

Does using rbind is the correct way? since I need to do another looping from SubInt.

Instead of `nrow(trainData)`, if you have `trainIndexes` and then use `w[SubInd, ]` at the end. If you have a proper example, it would have been easier to give a better answer. — kangaroo_cliff, Jul 17 '19 at 22:27

score 0 · Answer 1 · answered Jul 17 '19 at 23:08

If your only goal is to randomly sample 30% of your training data for each fold, then you can try using lapply() instead of the for-loop. Combined with filter() and sample_frac(). With 1000 original cases, the first fold's training data will have 900 cases, so 270 are returned when sampling 30%.

# create df
df <- data.frame(x=runif(1000))

#Randomly shuffle the data
df <- df[sample(nrow(df)),]; df <- data.frame(x=df)

#Create 10 equally size folds
folds <- cut(seq(1,nrow(df)),breaks=10,labels=FALSE)
df$folds <- folds

df1 <- lapply(1:10,function(i){
  train <- df %>% filter(folds!=i) %>% sample_frac(.3)
})

lapply(df1,dim)

d <- df1[[1]]; d %>% count(folds) # check no test data, fold==1
d <- df1[[2]]; d %>% count(folds) # check no test data, fold==2

score 0 · Answer 2 · answered Jul 18 '19 at 01:02

Since test folds are not of the same size, using c is easier to combine the resulting vectors.

I don't exactly know if this is what you are after. I am guessing it is either all_0.3train or the unique_0.3_train

for (i in 1 : kcv) { 

  trainIndexes <- which(kfolddata[, ncol(kfolddata)] !=i, arr.ind=TRUE)
  testData <- w[-trainIndexes, ]
  trainData <- w[trainIndexes, ]

  if (i == 1) {

    set.seed(1234)
    SubInd = sample(trainIndexes , size = round(0.3 * 
                                                nrow(trainData)), replace=F)
  } else {

    SubInd = c(trainIndexes, sample(trainIndexes , size = round(0.3 *
                                                  nrow(trainData)), replace=F))
  }
}

all_0.3train <- w[SubInd, ]
unique_0.3_train <- w[unique(SubInd), ]

Thank you @Suren. Nice output. – Norin Jul 18 '19 at 12:38 — Norin, Jul 18 '19 at 12:38

Select sample from train data based on fold from k-fold cross-validation

2 Answers2