autodiff fails on svd in pre_shape.py

[vontycowicz]

Dear geomstats team,

we are trying to perform geodesic regression in Kendall shape space but encountered the issue that the current implementation is not compatible with autodiff functionality.
In particular, the align method in geomstats/geometry/pre_shape.py employs singular value decomposition for which autodiff fails if a full set of left/right singular vectors are requested.
However, providing the flag 'full_matrices=False' avoids this pitfall and should yield the same alignment.

We added the flag and, indeed, are now able to run regression.
We will submit the modified pre_shape.py along with our project s.t. it does not rely on short notice updates of geomstats.

Best regards,
Christoph

[ninamiolane]

Thank you very much for the heads-up, and for finding this workaround. Looking forward to your submission!