Implement a from_records constructor

[grst]

I am missing a way to construct a direct way to construct a DataFrame from a list of tuples, as I would do in python with pandas.DataFrame.from_records.

Input data:

records = [("A", 1), ("B", 2), ("C", 3)]`

Desired DataFrame:

│ Row │ x1  │ x2 │
├─────┼─────┼────┤
│ 1   │ "A" │ 1  │
│ 2   │ "B" │ 2  │
│ 3   │ "C" │ 3  │

I now know that I can do something like

n = length(records[1])
DataFrame([[r[i] for r in records] for i=1:n],[Symbol("x$(i)") for i=1:n])

but this feels somehow cumbersome...

[quinnj]

I've wanted this as well.

[nalimilan]

Maybe we could implement this as DataFrame(::AbstractArray{<:Tuple}) constructor? Or could that conflict with other meanings of this signature?

[ararslan]

If I'm not mistaken, constructing a DataFrame from an array of arrays treats the interior arrays as columns, whereas in this proposal the tuples would be treated as rows. That's kind of weirdly inconsistent.

[grst]

How about simply supporting a DataFrame(::Zip) constructor?
Then we could do DataFrame(zip(records...)) which would not feel too awkward for me.

[nalimilan]

Good point. Special-casing Zip would be weird, but accepting any iterator of columns (including zip) would be a nice generalization of the existing DataTable(columns::Vector{Any}, names::Vector{Symbol}) constructor.

[nalimilan]

@quinnj How does this issue stand regarding the new Tables.jl constructors?

[bkamins]

We now allow it and generate column names with numbers as symbols due to Tabels.jl functionality.