I’d like to be able to create a row matrix holding vectors (the purpose is to pass them as series to Plots, but the question is general). However, I get issues with concatenation. I can create a vector of different-length vectors easily: a = [rand(n) for n in 2:5] but a' fails with an error relating to concatenating different length vectors. It does seem possible to create a 2-d array of arrays:

test = Matrix{Any}(5,5)
fill!(test, rand(5))

works, but only as long as the arrays all have the same length. Am I abusing the syntax?

You can skip the reshape (or transpose) call by using multi-dimensional comprehension syntax:

julia> [rand(n) for i=1:1, n=1:3]
1×3 Array{Array{Float64,1},2}:
[0.19346] [0.875024,0.573032] [0.26811,0.288699,0.481266]

Note that just using i=1 still only produces a 1-D vector since as a scalar it’s making a contribution of 0 to the overall result dimension.

Also, note that using fill! is probably wrong here because it’s only allocating and building a single vector of numbers and putting a reference to that same vector object in each place in the matrix.

should reshape a column vector to a row vector. Since reshape is a view it should be pretty cost-free too.

One place this comes up is when working with Plots.jl. A series is usually a column, and so rows denote the different series. So giving something like a labels for legends needs to be a row vector. I used to just transpose the vector of names (strings) I’d get in my recipes, but that’s being deprecated. The reshape trick works just fine though.