R split data into 2 parts randomly
Solution 1
Try
n <- 100
data <- data.frame(x=runif(n), y=rnorm(n))
ind <- sample(c(TRUE, FALSE), n, replace=TRUE, prob=c(0.7, 0.3))
data1 <- data[ind, ]
data2 <- data[!ind, ]
Solution 2
I am building on the answer by ExperimenteR, which appears robust. One issue however is that the sample
function is a bit weird in that it uses probabilities, which are not completely deterministic. Take this for example:
>sample(c(TRUE, FALSE), n, replace=TRUE, prob=c(0.7, 0.3))
You would expect that the number of TRUE
and FALSE
values to be exactly 70 and 30, respectively. Oftentimes, this is not the case:
>table(sample(c(TRUE, FALSE), n, replace=TRUE, prob=c(0.7, 0.3)))
FALSE TRUE
34 66
Which is alright if you're not looking to be super precise. But if you would like exactly 70% and 30%, then do this instead:
v <- as.vector(c(rep(TRUE,70),rep(FALSE,30))) #create 70 TRUE, 30 FALSE
ind <- sample(v) #Sample them randomly.
data1 <- data[ind, ]
data2 <- data[!ind, ]
gregorp
Updated on July 17, 2022Comments
-
gregorp almost 2 years
I am trying to split my data frame into 2 parts randomly. For example, I'd like to get a random 70% of the data into one data frame and the other 30% into other data frame. Is there a fast way to do this? The number of rows in the original data frame is over 800000. I've tried with a for loop, selecting a random number from the number of rows, and then binding that row to the first (70%) data frame using rbind() and deleting it from the original data frame to get the other (30%) data frame. But this is extremely slow. Is there a relatively fast way I could do this?
-
thelatemail almost 9 yearsBeat me to it. Works quickly when extended to 800K cases too.
-
gregorp almost 9 yearsWorks really fast. Even when I repeat it in a loop multiple times. Thank you.
-
moodymudskipper almost 6 yearsTry
ind <- sample(c(rep(TRUE,ceiling(nrow(data)*0.7)),rep(FALSE,floor(nrow(data)*0.3))))
-
Workhorse almost 6 yearsPartially true yes, ultimately needs numbers that are factors or multiple of 100. Yours is more robust, +1.
-
Workhorse almost 6 yearssorry meant to say my method ultimately needs numbers that are multiples of 10.