Generic square matrix constructor from a vector?

General Usage
1

Is there a generic way, given a vector, to produce a square matrix of the “best matching” type? By “best matching” I mean here Matrix for Base vectors and views, MMatrix for static arrays and SizedMatrix for SizedVectors from StaticArrays.jl.

For arrays in Base, there’s of course similar(vec, (length(vec), length(vec))). However, it does not produce the desired result with static vectors (returns an ordinary matrix for the sake of type stability).

I have come to the following:

function sqmatr(vec::AbstractVector{T}, F::DataType = T) where {T}
    cinds = CartesianIndices((eachindex(vec), eachindex(vec)))
    return similar(cinds, F)
end

which handles Base.Vector and SVector/MVector but not SizedVector (returns MMatrix when SizedMatrix would be desired).

The ideal option would be similar(vec * vec') but without the actual computation of the outer product.
Is there a way to do that without introducing an explicit dependency on StaticArrays.jl?

2

So, I eventually found the following solution:

function sqmatr(vec::AbstractVector{T}, F::DataType = T) where {T}
    axis, = axes(vec)
    return similar(vec, F, (axis, axis))
end
1 Like