No row/colSums for a DataFrame?

[LTLA]

I'm a bit surprised we don't have colSums and rowSums methods for DataFrame.

colSums(DataFrame(a=1:10))
## Error in base::colSums(x, na.rm = na.rm, dims = dims, ...) :
##  'x' must be an array of at least two dimensions
rowSums(DataFrame(a=1:10))
## Error in base::rowSums(x, na.rm = na.rm, dims = dims, ...) :
##   'x' must be an array of at least two dimensions

This should be as simple as:

setMethod("colSums", "DataFrame", function (x, na.rm = FALSE, dims = 1, ...) {
     # Some work required to respect 'dims', whatever that does.
     unlist(lapply(x, sum, na.rm=na.rm))
})

setMethod("rowSums", "DataFrame", function(x, na.rm=FALSE, dims=1, ...) {
     output <- integer(nrow(x))
     for (i in seq_along(x)) {
          y <- x[[i]]
            if (na.rm) {
              y[is.na(y)] <- 0
          }
          output <- output + y
     }
     output
})

[lawremi]

This was intentionally left out because those methods really only make sense on matrices. It's unfortunate that base::colSums() only coerces to matrix when x is a data.frame. Otherwise, it would just work, for better or worse.

I guess base::colSums() could coerce anything that has >= 2 dimensions to an array. One issue is that as.array() fails on a data.frame; it should probably just delegate to as.matrix().

[LTLA]

I guess base::colSums() could coerce anything that has >= 2 dimensions to an array.

That would make sense. I have some DataFrames of statistics that I just want to take colSums over, and it is annoying to have to remember to as.matrix() them when they should just work.

[LTLA]

Looking at base::colSums() indicates that it takes the time to check for array-ness anyway, so it should be a simple (and near cost-less) matter of just breaking up:

    if (!is.array(x) || length(dn <- dim(x)) < 2L) 
        stop("'x' must be an array of at least two dimensions")

and putting an as.matrix() call later under a lone !is.array(x) block.

[lawremi]

I think it should coerce to an array though, not necessarily a matrix, since the object could have more than two dimensions. The data.frame case could also be removed if as.array() gained a method to support data.frame.