Hermitian data fit

[aplavin]

Hi! This package looks interesting and pretty general, but I cannot find a way to fit a low rank model like A = X * X', where observed data A is a hermitian matrix (in my case - exactly hermitian), and Y is constrained to equal X'. Any ideas for how to proceed?

[madeleineudell]

This functionality does not yet exist. I can see two ways to implement it.

The simplest way is simply to create a new fit_hermitian! function. It could copy proxgrad.jl except that you'd replace the Y update by Y = copy(X).

A more complex way would create

• a new datatype HermitianGLRM extending AbstractGLRM; the main difference from the standard GLRM would be the absence of the Y field.
• a new algorithm to fit this model. It could be essentially the same as proxgrad.jl but you'd replace Y by X and remove the Y update.
• You might also need to add methods for a bunch of the convenience methods.

I don't plan to add this functionality, but would welcome a PR if the code lived entirely in new files (+ documentation in the README).