What's the best way to convert from Array{Any} to, say, Array{Int64}

[Marcos Sainz]

Besides the obvious looping answer, I'm using reduce as shown below when performance isn't key,

but is there are a more natural convert built into the API?

julia> arr = {[1,2] [2,4] [3,4]}

1x3 Any Array:

 [1,2]  [2,4]  [3,4]

julia> reduce(hcat, arr)

2x3 Int64 Array:

 1  2  3

 2  4  4

[Marcos Sainz]

And also, what's the best way to convert the reverse way.

Loops aside, I've got:

julia> reduce(hcat, Array{Any,1}, [1,2,3])

1x4 Any Array:

 Array{Any,1}  1  2  3 

[Marcos Sainz]

Oops, never mind for the reverse case:

julia> Any[1,2,3]

3-element Any Array:

 1

 2

 3

[Terry Seaward]

Hi Marcos, would something like this work?

 convert(Array{Int64}, Any[1 2 3; 4 5 6])

[Marcos Sainz]

Yes! Thank you Terry. 

[Marcos Sainz]

Related to the above:

What is the preferred way in julia to convert Array{Any} which is really an Array{Array{Float64}} into a Matrix{Float64}. For example:

julia> a

7-element Any Array:

 [1.0,1.0]

 [2.0,3.0]

 [4.0,7.0]

 [5.0,4.0]

 [7.0,2.0]

 [8.0,4.0]

 [9.0,6.0]

julia> reduce(hcat, a)

2x7 Float64 Array:

 1.0  2.0  4.0  5.0  7.0  8.0  9.0

 1.0  3.0  7.0  4.0  2.0  4.0  6.0

 Any better (i.e. more performant or idiomatic) way?

(sorry that I wasn't able to generalize the above answer to solve this case)

Thank you,

Marcos

[Simon Kornblith]

I think hcat(a...) should work for this.

[Marcos Sainz]

Thank you Simon. I can now stop abusing 'reduce'. 

How about to go in the other direction, Matrix{Float64} into a cell array of its rows (or columns)? What would be the preferred way?

[Simon Kornblith]

You should first consider if you need to convert the rows/columns of a matrix into a cell array, since it's rarely the most efficient way of doing things. If there's no better way to do what you want to do, you could use array comprehension:

{a[:, i] for i = 1:size(a, 2)} # cols
{a[i, :][:] for i = 1:size(a, 1)} # rows

or mapslices:

mapslices(x->{x}, a, 1)[:] # cols
mapslices(x->{x}, a, 2)[:] # rows

(The [:] above converts 2D row/column vectors into 1D column vectors. Depending on your application, you may not want/need this.)