I hope this question has not been asked before - but I could not find it:

Why does JacobiSVD not allow fixed column sizes when calculating thin unitaries? Especially when doing OLS columns are often fixed and rows are dynamic depending on sample count.

Is this a limitation of the current implementation or are there other reasons? As far as I can see fixed row size is working. I know I can calculate the SVD of the transposed but then I cannot use solve().

This is likely to be a minor bug, this should be fixed very soon. In the meantime use Dynamic for the column dimension.

hm.. actually that's not really possible because the thin/full status is provided at run-time while the dimension of fixed-size objects has to be known at compile-time.

Nevertheless I agree that needs to be fixed somehow...

Ah ok. Nice to know