Question

As per Row wise matrix operations in R I would like to apply a row-wise function on a data.table I have. I wish to calculate, per row, the mean of a number of columns in that row. My current attempt is:

columns <- c(1,5,10,15,20) # Actually obtained via grep
my.data.table[,"average" := mean(columns),with=FALSE] # Or...
my.data.table[,average := mean(columns)]

This, unfortunately, simply returns the mean of the ‘columns’ vector rather than mean of the columns to which they refer. Is there a way to refer to these columns by number?

Here’s the average I’m trying to achieve:

key  a b c average
A    5 5 5 5
B    1 2 3 2
C    2 4 9 5

Answer 1

Here are two possible solutions. They are basically both from the link that you’ve already provided, so maybe I missed something with this question. Here we go:

Solution 1 (using rowMeans):

library(data.table)
N <- 1000000
my.data.table <- data.table(ID = 1:N,
                            Year1 = rnorm(N),
                            Year2 = rnorm(N),
                            Year3 = rnorm(N),
                            Year4 = rnorm(N))

x <- c(2, 3, 4, 5)
system.time(x1 <- rowMeans(my.data.table[, ..x]))
   user  system elapsed 
   0.08    0.00    0.08

Solution 2: Get it into long format first. I thought this was faster, mainly because of Matthew’s comment in the other question that says that data.table is meant for the DT[,mad(variable),by=group] syntax. I think I’m missing something, but don’t see what:

library(reshape2)
DT <- as.data.table(melt(as.data.frame(my.data.table), id.var="ID"))
setkey(DT, ID)
system.time(x2 <- DT[, mean(value), by="ID"][[2]])
   user  system elapsed 
  11.28    0.00   11.33 
all.equal(x1, x2)
[1] TRUE