Splitting Column of a data.table

Question

Wonder how to split column of a data.table.

Can split column of a data.frame easily with the following code:

df <- data.frame(Test=c("A - B", "C - D"))
df
     Test
1 A - B
2 C - D
library(reshape2)
reshape2:::colsplit(string = df[,1], pattern = " ", names = c("Var1", "Space", "Var2"))

  Var1 Space Var2
1    A     -    B
2    C     -    D

But my attempt of splitting column of a data.table fails

library(data.table)
dt <- data.table(Test=c("A - B", "C - D"))
dt
    Test
1: A - B
2: C - D
reshape2:::colsplit(string = dt[,1, with=FALSE], pattern = " ", names = c("Var1", "Space", "Var2"))
Error: String must be an atomic vector

Answer 1

I see that you're specifically asking for something with colsplit, but I would suggest looking at some of the other alternatives out there, for example my cSplit function.

The cSplit approach would be as follows:

setnames(cSplit(dt, "Test",  " "), c("Var1", "Space", "Var2"))[]
#    Var1 Space Var2
# 1:    A     -    B
# 2:    C     -    D

The [] at the end is to print the result, but you can also just store the result as a new data.table instead.

---

How does it compare in terms of efficiency?

fun1 <- function() {
  reshape2:::colsplit(string = dt[[1]], pattern = " ", 
                      names = c("Var1", "Space", "Var2"))
} 
fun2 <- function() {
  setnames(cSplit(dt, "Test",  " "), 
           c("Var1", "Space", "Var2"))[]
}

dt <- rbindlist(replicate(5000, dt, FALSE))
dim(dt)
# [1] 10000     1

library(microbenchmark)
microbenchmark(fun1(), fun2(), times = 10)
# Unit: milliseconds
#    expr        min         lq     median         uq        max neval
#  fun1() 2025.84703 2093.39687 2195.75822 2390.30666 2492.65946    10
#  fun2()   34.08966   36.01145   43.28036   47.45962   57.57615    10

---

Why didn't your colsplit approach work as you expected?

dt[,1] is more like df[,1, drop = FALSE] (try it--you'll get the same error that you got with your "data.table" attempt).

You will need either of the following:

> dt[[1]]
[1] "A - B" "C - D"
> dt$Test
[1] "A - B" "C - D"

Which is similar to what you get with:

> df[, 1]
[1] A - B C - D
Levels: A - B C - D