derivative (L) matrix should have a square basis

[jjstickel]

It seems to me that the derivative L matrix should be square (mxm where m is the number of columns of A) for order 0 (i.e., square identity matrix) and decreasing rows for higher orders, rather than the shape of A. At least, that is what is needed for my data-smoothing application. I have made the necessary changes here:

jjstickel@604f359

Does this seem reasonable? Or are there cases when L should correspond to both dimensions of A? All the existing tests pass, but from what I can tell, they all use a square A. I'll make a PR later after I have worked out additional code I might want to add.

[mdpetters]

Thanks for the comment. After some testing with a non-square system I agree that L matrix should be square. In fact, the way it is implemented in the current version, L'L is missing terms in the last rows/columns, which leads to a corresponding lack or regularization at the edge of the domain. The changes mentioned above fix this, and thus also improve the package. A PR would be very welcome.

[jjstickel]

Great, thanks for looking at this. I'll submit a PR in the next few days. Still working on some other concepts that may go in this PR or in a second one at a later time.