I also ended up removing the global invretract since I think it should be specific to Stiefel/Grassmann/Unitary, just like e.g. project. I guess that makes this a breaking change, at least in principle.
Also, I tried adding a test for Stiefel.invretract too, but it errored in _stiefellog with
AssertionError: mapreduce(abs ∘ imag, max, logU; init = abs(zero(eltype(logU)))) <= tol
I could look into this, but maybe you know what's going on already?
I also ended up removing the global
invretract
since I think it should be specific to Stiefel/Grassmann/Unitary, just like e.g.project
. I guess that makes this a breaking change, at least in principle.Also, I tried adding a test for
Stiefel.invretract
too, but it errored in_stiefellog
withAssertionError: mapreduce(abs ∘ imag, max, logU; init = abs(zero(eltype(logU)))) <= tol
I could look into this, but maybe you know what's going on already?