Extend CovarianceEstimation.cov

[glennmoy]

The current covariance (and correlation) methods overload Statistics.cov (Statistics.cor).

When using CovarianceEstimation.jl another cov method is exported that takes a CovarianceEstimator type.

The issue is that as of 0.2.4, cov returns a Symmetric matrix which incorrectly wraps the NamedDimsArray if it gets given one.

julia> using CovarianceEstimation, LinearAlgebra, NamedDims

julia> x = NamedDimsArray{(:a, :b)}(rand(5, 5));

julia> estimator = LinearShrinkage(DiagonalCommonVariance(); corrected=true)

julia> cov(estimator, x)
5×5 Symmetric{Float64,NamedDimsArray{(:b, :b),Float64,2,Array{Float64,2}}}:
  0.0668947  -0.0         0.0         0.0        -0.0
 -0.0         0.0668947  -0.0         0.0         0.0
  0.0        -0.0         0.0668947  -0.0         0.0
  0.0         0.0        -0.0         0.0668947  -0.0
 -0.0         0.0         0.0        -0.0         0.0668947

We need to extend the function properly with:

function CovarianceEstimation.cov(
    estimator::LinearShrinkage, a::NamedDimsArray{L, T, 2};
    dims=1,
) where {L, T}
    numerical_dims = dim(a, dims)
    data = cov(estimator, parent(a); dims=numerical_dims)
    names = symmetric_names(L, numerical_dims)
    return NamedDimsArray{names}(data)
end

then we get

julia> cov(LinearShrinkage(DiagonalCommonVariance(); corrected=true), x)
5×5 NamedDimsArray{(:b, :b),Float64,2,Symmetric{Float64,Array{Float64,2}}}:
  0.0668947  -0.0         0.0         0.0        -0.0
 -0.0         0.0668947  -0.0         0.0         0.0
  0.0        -0.0         0.0668947  -0.0         0.0
  0.0         0.0        -0.0         0.0668947  -0.0
 -0.0         0.0         0.0        -0.0         0.0668947