Antoine Collas

@nkoep I fixed the issue with the `random_point` method by creating the metaclass `BackendManifold` that wraps `random_point`.

I'll try to get an example like `dominant_eigenvector.py` working end to end with `torch` to get a proof of concept.

@nkoep, I changed quite a lot of things, but we are getting close to having the first version of the multi-backend support for manifolds operations! For now, I focused on...

Hi, Can you give a minimal reproducible example so that we can see what the problem is in detail? For example, you could adapt the following example with numba. Be...

I agree with @sweichwald . We must handle this "backend everywhere" such that we have low maintenance and easy entry for newcomers. What @nkoep has proposed in #246 seems to...

What @nkoep proposed in #246 are general manifolds, i.e., given a cost function defined with a specific backend, the chosen manifold can perform the operations on this backend. But I...

Hello, indeed, the set of ${w \in \mathbb{C}^n: || w || = 1\}$ is a manifold. Unfortunately, it is only available as a real-valued manifold in Pymanopt. If you implement...

All the backend (except maybe tensorflow) compute on the fly. So there is no need to tell the manifold which backend it will use in advance.

By the way, if the goal is only to do optimization on the complex sphere manifold, I think it is better to directly implement the complex Stiefel manifold (which reduces...

I updated the PR and it passes all the tests now ! For me it is ready to be merged. What do you think @nkoep ?