(F)PSWF Performance and Validation

[garrettwrong]

Researcher has requested that we review the performance, validity, and state of integration of the (F)PSWF codes. There are a few open issues on this topic from before I joined which I will tag to this issue. My recollection is that these components are not well tested in Python.

I already have a branch (#515) where I have stubbed in changes to speak Image and refactored FPSWF to use nufft's many feature.

As I understand it, researcher goal is to use FPSWF in place of FFB in cov2d. This looks to involve reconciling the basis classes/implementations (including indices/grids), reconciling basis interpretations of CTF, and potentially reconciling some covar calculations that seem tied to (F)FB. Thinking this is at least a moderately sized project.

Before I can estimate time to refactor the basis classes I need to understand the (F)PSWF better.

[garrettwrong]

#67 #74 #73

[janden]

If I understand correctly, the PSWF and FB have the same angular components but differ in their radial components. As a result, a lot of the implementation of (F)FBBasis can be carried over to F(PSWF)Basis. The problem is that they're structured very differently and have different outputs (as you mention). So in theory, this is all possible, but it's going to be a significant amount of work (couple of thousand lines of code). If this is a priority, we can discuss in the meeting.