This was suggested by @alasfar-lina.
Let's say you have several best fit vectors with shape (N, K...) and covariance matrices with shape (N, N, K...), where N is the number of parameters, and K... are other dimensions. Then Lina wants propagate_covariance to transform all of them at once. This could be implemented perhaps with a simple change to the internal logic (to be confirmed).
Question is, should we also allow (K..., N) vectors and (K..., N, N) matrices as input? How can the two cases be distinguished in the call? Can we learn from numpy/scipy interfaces on how to do this?
This was suggested by @alasfar-lina.
Let's say you have several best fit vectors with shape
(N, K...)and covariance matrices with shape(N, N, K...), where N is the number of parameters, andK...are other dimensions. Then Lina wantspropagate_covarianceto transform all of them at once. This could be implemented perhaps with a simple change to the internal logic (to be confirmed).Question is, should we also allow
(K..., N)vectors and(K..., N, N)matrices as input? How can the two cases be distinguished in the call? Can we learn from numpy/scipy interfaces on how to do this?