I currently have to handle multi-dimensional data, where I have to compute several broadcasting operations with ranges.
In 2D I use e.g (1:10) .* (1:5)' which gives a nice matrix.
The first vector is of type UnitRange{T} and the second results in a Adjoint{T,UnitRange{T}} (with here T = Int64).
What would be the recommended way to broadcast theses ranges to higher dimensions ?
I thought about initializing a matrix of singleton dimensions except one then fill it with a range, but it feels a bit weird.
For instance:
ndrange = zeros(1,1,1,10,1)
ndrange[1,1,1,:,1] = 1:10
You can use reshape(1:10, 1, 1, 1, 10, 1) for broadcasting, but depending on what you’re doing, there may be better options (plain loops, mapslices, JuliennedArrays, Tullio).
2 Likes
As @stillyslalom said, I would appeal to something beyond the usual dot-broadcasting for this. mapslices or Tullio.jl are great alternatives.
1 Like
It might be easier and more efficient to just have range-like higher dimensional arrays. E.g., the rather barebones
1 Like
Another perhaps relevant package here, which generalises permutedims (the backend of a reworking of TensorCast.jl):
julia> using TransmuteDims
julia> transmutedims('a':'e', (2,1)) # makes an Array, knows that size('a':'e',2)==1
1×5 Array{Char,2}:
'a' 'b' 'c' 'd' 'e'
julia> transmute(10:10:30, (3,2,1)) # lazy version, notice (3,2,1) equivalent to (0,0,1)
1×1×3 transmute(::StepRange{Int64,Int64}, (0, 0, 1)) with eltype Int64:
[:, :, 1] =
10
[:, :, 2] =
20
[:, :, 3] =
30
1 Like
Well I guess, Tullio would probably be very useful! I will have a look at it ! Thanks everyone for the help