apply for sparseMatrix in R

Question

I am wondering is there a way to perform some kind of apply function on a sparseMatrix (from Matrix package) in R to cut Columns on k equinumerous groups?

And is there away to divide for groups only those elements in column that are greater than 0?

For small sparseMatrix code looks like this but I bet it won't work efficient on bigger matrix.

library(Matrix)
i <- c(1:8, rep(8,7)); j <- c(1:8, 1:7); x <- c(8 * (1:8),1:7)
(A <- sparseMatrix(i, j, x = x))
#8 x 8 sparse Matrix of class "dgCMatrix"

[1,] 8  .  .  .  .  .  .  .
[2,] . 16  .  .  .  .  .  .
[3,] .  . 24  .  .  .  .  .
[4,] .  .  . 32  .  .  .  .
[5,] .  .  .  . 40  .  .  .
[6,] .  .  .  .  . 48  .  .
[7,] .  .  .  .  .  . 56  .
[8,] 1  2  3  4  5  6  7 64
> 
"
> k<- 2
> apply(A,2,function(element){
+   cut(element,
+   k)})
     [,1]         [,2]         [,3]          [,4]          [,5]         [,6]          [,7]          [,8]         
[1,] "(4,8.01]"   "(-0.016,8]" "(-0.024,12]" "(-0.032,16]" "(-0.04,20]" "(-0.048,24]" "(-0.056,28]" "(-0.064,32]"
[2,] "(-0.008,4]" "(8,16]"     "(-0.024,12]" "(-0.032,16]" "(-0.04,20]" "(-0.048,24]" "(-0.056,28]" "(-0.064,32]"
[3,] "(-0.008,4]" "(-0.016,8]" "(12,24]"     "(-0.032,16]" "(-0.04,20]" "(-0.048,24]" "(-0.056,28]" "(-0.064,32]"
[4,] "(-0.008,4]" "(-0.016,8]" "(-0.024,12]" "(16,32]"     "(-0.04,20]" "(-0.048,24]" "(-0.056,28]" "(-0.064,32]"
[5,] "(-0.008,4]" "(-0.016,8]" "(-0.024,12]" "(-0.032,16]" "(20,40]"    "(-0.048,24]" "(-0.056,28]" "(-0.064,32]"
[6,] "(-0.008,4]" "(-0.016,8]" "(-0.024,12]" "(-0.032,16]" "(-0.04,20]" "(24,48]"     "(-0.056,28]" "(-0.064,32]"
[7,] "(-0.008,4]" "(-0.016,8]" "(-0.024,12]" "(-0.032,16]" "(-0.04,20]" "(-0.048,24]" "(28,56.1]"   "(-0.064,32]"
[8,] "(-0.008,4]" "(-0.016,8]" "(-0.024,12]" "(-0.032,16]" "(-0.04,20]" "(-0.048,24]" "(-0.056,28]" "(32,64.1]"  

Answer 1

Three possible approaches:

1. Convert the sparse matrix to a data.table
2. Convert to a simple_triplet_matrix and use the rollup function from the slam package.
3. Convert the sparse matrix to a list of columns and use vapply

Options 1 and 3 support parallel processing over columns. Option 3 has the fewest dependencies. An implementation of option 3 is available as part of the quminorm package. If I find time I might spin it off into a separate package in the future. Note that for functions that also require the zero values the best approach is to use function colapply_simple_triplet_matrix from package slam.

Here is a vignette comparing a variety of different schemes in terms of speed and memory consumption.