Here is a 10 x 12 matrix:
mat <- matrix(runif(120, 0, 1), 10)
I am trying to find column sums for subsets of a matrix (specifically, column sums for columns 1 through 4, 5 through 8, and 9 through 12) by row. The desired output would be a 10 x 3 matrix.
I tried the approaches from this answer using tapply and by (with detours to rowsum and aggregate), but encountered errors with all of them.
What the OP is describing is called a row sum in R:
# using Matthew Lundberg's example data
x <- matrix(1:36, 3,12)
g = split(seq(ncol(x)), (seq(ncol(x)) - 1) %/% 4 )
sapply(g, function(cols) rowSums( x[, cols] ))
# 0 1 2
# [1,] 22 70 118
# [2,] 26 74 122
# [3,] 30 78 126
It's typical to have grouping variables over rows/observations not columns/variables. To reach this case, the OP could transpose:
rowsum( t(x), (seq(ncol(x))-1) %/% 4 )
# [,1] [,2] [,3]
# 0 22 26 30
# 1 70 74 78
# 2 118 122 126
You can do this with a brute-force approach, specifying each column within apply:
t(apply(x, 1, function(y) c(sum(y[1:4]), sum(y[5:8]), sum(y[9:12]))))
It's easier to see with non-random data, and a shorter matrix for input:
> x <- matrix(1:36, 3,12)
> x
[,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10] [,11] [,12]
[1,] 1 4 7 10 13 16 19 22 25 28 31 34
[2,] 2 5 8 11 14 17 20 23 26 29 32 35
[3,] 3 6 9 12 15 18 21 24 27 30 33 36
> t(apply(x, 1, function(y) c(sum(y[1:4]), sum(y[5:8]), sum(y[9:12]))))
[,1] [,2] [,3]
[1,] 22 70 118
[2,] 26 74 122
[3,] 30 78 126
You can also split the vector with split, and while this is more idiomatic for R and more flexible, it is not really more readable:
> t(apply(x, 1, function(y) sapply(split(y, ceiling(seq_along(y)/4)), sum)))
1 2 3
[1,] 22 70 118
[2,] 26 74 122
[3,] 30 78 126
We could convert to array, use apply with MARGIN=1 and get the colSums
n <- 4
t(apply(array(mat, dim=c(nrow(mat), n, ncol(mat)/n)), 1, colSums))
Or another option is melt/acast from library(reshape2)
library(reshape2)
acast(melt(mat), Var1~(Var2-1)%/%n, value.var='value', sum)
The wrapper function recast can be used to make this compact
recast(mat, Var1~(Var2-1)%/%4, id.var=NULL, sum)
