Antoine Collas

Also, I struggled a bit with the pull requests and I closed the other pr I created (#119 #121 #122).

Yes it supersede #119 #121 #122 !

This complex implementation already has a multitransp. Normally, the inner product is fine !

@nkoep I implemented the numpy backend for all the functions you added to the TODO list.

@nkoep `examples.closest_unit_norm_column_approximation` is now working with the numpy backend! However, we have a problem when calling the method `random_point` of a given manifold. It uses the numpy backend because it...

The last commit is an attempt to alleviate this problem. By default `random_point` generates a `numpy.ndarray`. However, if the attribute `manifold.backend` is set to something else (pytorch, jax or tensofrflow),...

@nkoep tell me if it is ok for you. Then, I'll fix the tests that now fail, and I'll implement the `pytorch` backend.

Locally, all tests are passing now!

The problem is that 'random_point' does not know which backend it should use. Comparatively, you give a 'point' to 'random_tangent_vector' so it knows which backend it has to use.

Thanks for all the comments Niklas! I'll go through them in the coming week.