Closed kellertuer closed 3 months ago
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 99.58%. Comparing base (
6f54b6e
) to head (36a00d1
).
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
That was quick! The failure is caused by a new ambiguity:
injectivity_radius(M::AbstractDecoratorManifold, m::AbstractRetractionMethod) @ ManifoldsBase ~/.julia/packages/ManifoldsBase/pOL51/src/nested_trait.jl:305,
injectivity_radius(::Stiefel, p) @ Manifolds ~/work/Manifolds.jl/Manifolds.jl/src/manifolds/StiefelEuclideanMetric.jl:112
Oh we should maybe check ambiguities on the infectivity radius. – So how do we reduce this? Or do we increase the bound?
It's real, you can try calling injectivity_radius(Stiefel(5, 2), ExponentialRetraction())
and it would fail. You can probably just delete injectivity_radius(::Stiefel, p) = π
though.
Then the special point tests fail. But if that is ok, we can do that (that is remove the point ones).
Then the special point tests fail. But if that is ok, we can do that (that is remove the point ones).
You can limit the type of p
in injectivity_radius(::Stiefel, p)
to those special point types.
or leave them out, we would specialise to arrays or such, and that might not be to useful (we otherwise do not do that I think).
I think the variant with p
should work. You can also disambiguate by defining injectivity_radius(M::Stiefel, m::AbstractRetractionMethod)
. Or we shouldn't have made injectivity_radius
so prone to ambiguities.
I omitted the p variants for now, but sure in the long run we should make the infectivity radius a bit less prone to ambiguities.
cf. Zimmermann & Stoye https://arxiv.org/abs/2405.02268v1 .