Nicolas Guigui

This does not concern the Biinvariant metric only but all invariant metrics. However they must take the `group` as input parameter, and currently `dim, point_type` are deduced from the `group`'s...

That'd be neat! It might not be possible for the quotient though, because one needs to specify the ambient metric on the total space.

Hi Emile, thank you for your comment! Could tell me more about which group your are looking at and which metric? There are different aspects to your question: - In...

Hi all, Thank you Luis for the proposal, here are my thoughts after reading the above discussion. - I was surprised by the name `GammaGeodesicVPSolver` because solvers should be general...

That's probably why we always have to adapt tol for tests in pytorch and tensorflow! I would have said float32 but we still need to gain on numerical accuracy...

We didn't need it so far, but for completeness a basis should be implemented in `Euclidean` and it will be inherited in Minkowski

This has been partially done manually. Could-it be automated @SaitejaUtpala ?

I do not agree, `group` is necessary in every `InvariantMetric` for computations (e.g. tangent_translation) more than for parameters (like dim, point_type).

It really depends on the numerical precision used though. Take a matrix with eigen value 1e-6, and try the belongs method on it - with numpy backend, it would return...

Hi! Thank you for spotting this, there is indeed a particular case for the von Mises-Fisher distribution on the circle, [this paper](https://www-jstor-org.proxy.unice.fr/stable/2346732?seq=1#metadata_info_tab_contents) could be used to implement this particular case.