Equivalent of invretr in matlab manopt in pymanopt

pymanopt / pymanopt

Python toolbox for optimization on Riemannian manifolds with support for automatic differentiation

https://pymanopt.org

BSD 3-Clause "New" or "Revised" License

761 stars 150 forks source link

Equivalent of invretr in matlab manopt in pymanopt #210

Open jameschapman19 opened 2 years ago

jameschapman19 commented 2 years ago

Trying to find the equivalent of the inverse retraction available in matlab manopt manifold factories for pymanopt manifolds? Any hints?

kellertuer commented 2 years ago

In the manifolds I checked I could not find such a method – which manifold do you plan to use? We (you start I help :)) could get a PR for some first retractions going If you like?

jameschapman19 commented 2 years ago

First! I should have added great work on this package :)

On inverse retraction a couple of examples (though I must admit I didn't know it wasn't a standard method in all the MATLAB ones because it just happened to be in the ones I had seen!):

I happen to be interested in https://github.com/NicolasBoumal/manopt/blob/master/manopt/manifolds/grassmann/grassmannfactory.m
Another example https://github.com/NicolasBoumal/manopt/blob/master/manopt/manifolds/hyperbolic/poincareballfactory.m but in this one it seems quite simple because M.retr = M.exp and M.invretr = M.log;

jameschapman19 commented 2 years ago

In principle I'm also happy to take a look at implementing the invretr myself for the ones I'm interested in and can put in a PR if it works for me

kellertuer commented 2 years ago

Oh, I do not work (much) on this package – maybe I have fixed a bug here and there – I develop Manopt.jl and hence know the structure and could easily check for the functions.

Yeah for the first case maybe check and implement a inverse_retraction in the Grassmann manifold class.

For the second, just use exp/log then (I am not sure why Manopt sets them that way then).