Base.unzip()

[ZacCranko]

Hi there,

apologies if this has already been addressed somewhere, but is there a reason that there is no unzip() function in Base?

Ideally this would be a function that would take a Vector{Tuple{ ... }} and return a Tuple{Vector, ..., Vector} for output. E.g.

julia> v = [(1,"a",:meow), (2,"b",:woof), (3,"c",:doh!)]; unzip(v)
([1,2,3],ASCIIString["a","b","c"],[:meow,:woof,:doh!])

A naive implementation might be something like

function unzip(input::Vector)
    n = length(input)
    types  = map(typeof, first(input))
    output = map(T->Vector{T}(n), types)

    for i = 1:n
       @inbounds for (j, x) in enumerate(input[i])
           (output[j])[i] = x
       end
    end
    return (output...)
end

[StefanKarpinski]

We definitely need an unzip function in Base.

[ivarne]

Or even simpler

unzip(a) = zip(a...) 

[yuyichao]

No

[StefanKarpinski]

Trying the zip(...) thing on any non-toy problem shows that it's not so good pretty fast, unfortunately.

[tlnagy]

I just ran into this issue when I absentmindedly used the Python-esque zip(arr...) where length(arr)=6000. It was not pretty. A dedicated unzip would be nice. Maybe even throw a warning that you shouldn't splat anything nontrivial.

[yuyichao]

Maybe even throw a warning that you shouldn't splat anything nontrivial.

Like almost all other performance tips in julia, doing so is perfectly fine and sometimes wanted for various reasons. (OK, splatting 6000 elements is a bit crazy, but splatting unknown size can be useful....). The only thing that matters is to not put it in performance critical path.

It looks like this is not covered in the performance tip?

[tlnagy]

Like almost all other performance tips in julia, doing so is perfectly fine and sometimes wanted for various reasons. (OK, splatting 6000 elements is a bit crazy, but splatting unknown size can be useful....). The only thing that matters is to not put it in performance critical path.

Yeah, a warning is a bad idea.

It looks like this is not covered in the performance tip?

I think this should also be marked under differences from other languages. People coming from Python are probably used to the zip(*arr) syntax and zip(arr...) might be a tempting target.

[hayd]

You can dispatch only on arrays of tuples:

function unzip{T<:Tuple}(A::Array{T})
    res = map(x -> x[], T.parameters)
    res_len = length(res)
    for t in A
        for i in 1:res_len
            push!(res[i], t[i])
        end
    end
    res
end

[StefanKarpinski]

Bump, we still need and unzip function.

[StefanKarpinski]

BUMP if anyone is bored...