Both of these can be efficiently checked for positive (semi) definiteness and would seem to fit here. There seem to be multiple instantiations, though: should the existing PDMat be renamed PDCholesky? Or should we add a F<:Factorization{T} parameter? Either is workable, I'm opening this largely to ask for guidance about your preferences.
We are about to unify PDMat and PDSparseMat in #188. This will already introduce a F<:Factorization type parameter, so extending PDMats to other factorizations will be much simpler after this change.
Based on https://github.com/mateuszbaran/CovarianceEstimation.jl/issues/90 we're likely to want something like
PDMat
that supports different factorizations. Two factorizations are currently under discussion:Eigen
SymWoodbury
with aDiagonal
orUniformScaling
"main matrix" (A
in https://en.wikipedia.org/wiki/Woodbury_matrix_identity)Both of these can be efficiently checked for positive (semi) definiteness and would seem to fit here. There seem to be multiple instantiations, though: should the existing
PDMat
be renamedPDCholesky
? Or should we add aF<:Factorization{T}
parameter? Either is workable, I'm opening this largely to ask for guidance about your preferences.CC @mateuszbaran