CODEWITHSUNDEEP
multidimensional-arrayjulia
Phuoc
Phuoc

Reputation: 1103

How to convert a multidimensional array to/from vector of vector of ... vector in julia

Is there a method in julia to convert a multidimensional array to a vector of vector and so on, and vice versa? It is OK to define a method for a fix number of dimensions. But how about a method for arbitrary dims?

julia> s = (1,2,3)

julia> a = reshape(1:prod(s), s)
1×2×3 Base.ReshapedArray{Int64,3,UnitRange{Int64},Tuple{}}:
[:, :, 1] =
 1  2

[:, :, 2] =
 3  4

[:, :, 3] =
 5  6

julia> b = [[[a[i,j,k] for i=1:s[1]] for j=1:s[2]] for k=1:s[3]]
3-element Array{Array{Array{Int64,1},1},1}:
 Array{Int64,1}[[1], [2]]
 Array{Int64,1}[[3], [4]]
 Array{Int64,1}[[5], [6]]

julia> unstack(a) == b
ERROR: UndefVarError: unstack not defined

Upvotes: 0

Views: 251

Answers (1)

Chris Rackauckas
Chris Rackauckas

Reputation: 19162

RecursiveArrayTools.jl can help with this kind of work. 

recs = [rand(8) for i in 1:10]
A = VectorOfArray(recs)
A[i] # Returns the ith array in the vector of arrays
A[j,i] # Returns the jth component in the ith array
A[j1,...,jN,i] # Returns the (j1,...,jN) component of the ith array

So it acts like the matrix without ever building the matrix, which is a good way to save allocations if you tend to act on the columns (which are the separate arrays). It also has a fast conversion to a contiguous array via the indexing fallback (honestly, I tried to create a faster one but the fallback worked better than I could make it):

arr = convert(Array,A)

Converting back would require allocating of course

VA = VectorOfArray([A[:,i] for i in size(A,2)])

Upvotes: 2

Related Questions