I'm experimenting with the new RleArrays from DelayedArray, and I was happy to see most operations work seamlessly just like a normal matrix. 

I often use the the very fast rowsum function to aggregate rows of matrix, but this does not seem to work on an RleArray. Is there an alternative to the rowsum-function in for RleArrays?

The closest I can think of is some work by Peter Hickey: https://github.com/PeteHaitch/DelayedMatrixStats. I don't know if this is applicable to RleArray objects, though.

Hi,

No rowsum method for DelayedMatrix objects yet, but there is a rowSums method. In the meantime, here is a quick and dirty implementation for rowsum that doesn't support reorder or na.rm (like rowsum.default and rowsum.data.frame do):

rowsum.DelayedMatrix <- function(x, group, reorder=TRUE, ...)
    apply(x, 2, function(col) sum(splitAsList(col, group)))

Then:

library(DelayedArray)
M <- as(matrix(runif(100000), nrow=2000), "RleArray")
group <- sample(8, 2000, replace=TRUE)
identical(rowsum(M, group), rowsum(as.matrix(M), group))
# [1] TRUE

Is not as fast as on an ordinary matrix though...

Cheers,

H.

If a DataFrame of Rles is good for you, you might be interested to have a look at my benchmark, where I concluded that rowsum(as.data.frame(lapply(DF, decode)), group) is the fastest option.