Constructor behavior on nested array vs array of tuple

[darrencl]

Hi,

I notice there is different behavior in constructing over a nested array vs array of tuple.

julia> DataFrame([[1,2,3],[4,5,6]])
3×2 DataFrame
│ Row │ x1    │ x2    │
│     │ Int64 │ Int64 │
├─────┼───────┼───────┤
│ 1   │ 1     │ 4     │
│ 2   │ 2     │ 5     │
│ 3   │ 3     │ 6     │

julia> DataFrame(Tuple.([[1,2,3],[4,5,6]]))
2×3 DataFrame
│ Row │ 1     │ 2     │ 3     │
│     │ Int64 │ Int64 │ Int64 │
├─────┼───────┼───────┼───────┤
│ 1   │ 1     │ 2     │ 3     │
│ 2   │ 4     │ 5     │ 6     │

Converting each element of array to tuple is quite slow (with list comprehension) and broadcasting (like the sample above) in large inner-arrays is prone to stack overflow error.

I am just wondering what are the alternatives to get the same behavior is array of tuple without the need of converting each element to tuple?

Thanks!

[bkamins]

I notice there is different behavior in constructing over a nested array vs array of tuple.

The difference is that the first example follows a DataFrame constructor and the second is a conversion from a type supported by Tables.jl API to a DataFrame.

The second operation must be slower than the first one. But what could be potentially faster if you have many rows is:

DataFrame!([getindex.(src, i) for i in 1:maximum(length.(src))]);

(this will also check if all your inner vectors have the same length)