Broadcast over matrix dimension for multiple arguments

General Usage
1

Say I have 2 matrices, a and b. How can I call a function that expects 2 vectors by broadcasting over their (eg) rows. For example:

function foo(a :: Vector, b :: Vector, c :: Float64)
    return (sum(a) + sum(b)) / c
end

a = [1 2 3 4; 5 6 7 8]
b = [1 2 3 4; 5 6 7 8]
c = 2.0

Should return a Vector of [10, 26]. Something that does not work but points to how this could be done is using foo.(eachrow(a), eachrow(b), c). Or similarly, mapslices(a->foo(a,b), a, dims=2), even though it does not work either since b is not sliced into vectors.

2 
julia> function foo(a, b, c)
           return (sum(a) + sum(b)) / c
       end

julia> foo.(eachrow(a),eachrow(b),c)
2-element Vector{Float64}:
 10.0
 26.0

btw, you probably should drop any and all type annotation, the c::Float64 is clearly too restrictive, and the minimal fix you need is Vector → AbstractVector

2 Likes
4

It does work, but you sabotage the function by putting in the overly restrictive type annotations.

5

Ah right, thanks for clarifying!

6

You can try AbstractVector, or just leave it untyped. Also, Float64 is much more specific than needed. Maybe use Number.

7

Yeap. Actually I have it now annotated as foo(a :: Vector{T}, b :: Vector{T}, c :: T) where T <: AbstractFloat, which works and it is more case specific for me.

8

That should not work, since eachrow does not return Vector. Perhaps you have some old definitions there.

1 Like
9

Spot on… thanks again!