Missing support for PEigenDecomposition core.matrix protocol

[jlowenz]

It looks like Eigen and IEigenResult are included from the vectorz implementation, but the core.matrix.protocols.PEigenDecomposition protocol is not implemented. Any thoughts on an ETA? Thanks!

[mikera]

Implementing the protocols should be easy enough, I can do that pretty quickly. I might need some help testing though - I'm not an expert on what the right test cases would be!

[jlowenz]

Ok, thanks, that sounds good. I might be able to help a bit on the testing - not a linalg expert myself though! One complication is that the eigenvalues and eigenvectors may be complex, even for real (non-symmetric) matrices... This seems like it might require core.matrix.complex to be present; what do you think?

[Torvaney]

Would it be possible/reasonable to expose the existing (commented out) implementation of eigen decomposition? Perhaps with a warning on complex numbers.

I can understand wanting to get it right, but as far as I'm aware, the majority of use cases are using real numbers.

[winkpoke]

Agree with @Torvaney. The eigen decomposition is commented out and completely not functioning now.