scaled vs canonical

[stephens999]

I wonder if maybe we should refer to the "scaled" matrices as "canonical" to be consistent with mashr
(and mvsusieR). Or at least use canonical to refer to the matrices that are being scaled.
eg in ud_init the U_scaled parameter would become U_canonical?

(Is it too confusing to use scaled.update to refer to the updates that estimate the scaling of the canonical matrices?)

[pcarbo]

Personally I find "scaled" much more clear than "canonical", though I understand the interest in being consistent.
I am okay with this if the scaled matrices will primarily be canonical matrices.
@yunqiyang0215?

[pcarbo]

Also "scaled" is more consistent with how we name the other matrices as "unconstrained" and "rank-1", which alludes to how these matrices are parameterized.

[stephens999]

I do agree that "scaled" is more consistent with "unconstrained".

If we are estimating U_k = sU_0 then I would say that U_0 is the "canonical" matrix and
U_k is a scaled (canonical) matrix.

[stephens999]

maybe we can keep scaled here and make the connection with canonical matrices in the documentation....

[yunqiyang0215]

Maybe we keep "scaled.update" for now and make a connection between "scaled" and canonical covariance matrices in the documents? And I agree that U0 is a canonical matrix and U (=s*U0) is a scaled matrix. Scaled/unconstrained/rank1 updates are consistent.