Summing DelayedArray and Matrix

[const-ae]

Hi Hervé,
I have recently started to play around with the DelayedArray package and it is just great to handle datasets that are too big to be kept in memory.

However, I recently tried to sum a classical matrix with a DelayedArray of the same dimension. To my surprise this gave the following error:

suppressPackageStartupMessages({
  library(DelayedArray)
})
# Standard Matrix
mat1 <- matrix(1:12, nrow = 3, ncol = 4)
mat2 <- matrix(3, nrow = 3, ncol = 4)
mat1 + mat2
#>      [,1] [,2] [,3] [,4]
#> [1,]    4    7   10   13
#> [2,]    5    8   11   14
#> [3,]    6    9   12   15

# Operations with DelayedArray
dmat1 <- DelayedArray(mat1)
dmat2 <- DelayedArray(mat2)
dmat1 + dmat2
#> <3 x 4> DelayedMatrix object of type "double":
#>      [,1] [,2] [,3] [,4]
#> [1,]    4    7   10   13
#> [2,]    5    8   11   14
#> [3,]    6    9   12   15

# Mixed
dmat1 + mat2
#> Error in .check_Ops_vector_arg_length(e, x_nrow, e_what = e_what, x_what = x_what): right object is longer than first dimension of left object

^{Created on 2020-01-23 by the reprex package (v0.3.0)}

My question is: is this the intended behavior?

I realize that this is not the most realistic scenario because if dmat1 is too big for the memory, mat2 would probably be as well. However, I expected this to work because now, if I write a generic function, I will have to check the classes of the input arguments and make sure that they match.

[hpages]

Hi Constantin,

The error is expected but the error message is not really helping.

From a technical point of view arithmetic operations between DelayedArray and non-DelayedArray operands are actually only supported when the non-DelayedArray operand is a "small" vector. The typical situations are:

x +   5  # add single value
x + 1:2  # add small vector (after recycling it)

"small" here means that the length of the vector should not be longer than the 1st dimension of x. Another restriction is that its length must be a divisor of the 1st dimension of x.

For convenience, I also added support for ordinary matrices or arrays but they're basically treated as vectors (they go thru as.vector()). This explains the error message that you get.

I've tried to improve the error message a little bit in DelayedArray 0.13.4 (commit cdfd4d6):

> dmat1 + mat2
Error in .check_Ops_vector_arg_length(e, x_nrow, e_what = e_what, x_what = x_what) : 
  when the right operand is not a DelayedArray object (or derivative),
  its length (12) cannot be greater than the first dimension of the left
  operand (3)

H.

[const-ae]

Thanks, for the explanation that makes sense. The error message is much more helpful now :)