FISTA iterative reconstruction method

[dkazanc]

Implement FISTA as a part of TomoPhantom.

[paskino]

I guess that's a bit of a political question, but why would you do it and not use the cil implementation?

As far as I can tell it would work fine on numpy arrays provided that you passed appropriate functions

[dkazanc]

Classical FISTA it's just a few lines, so it almost done. Why not CIL, a number of reasons.

• Need a quick access implementation in order to modify, test and upgrade things. Will be changing/tweaking things to accelerate FISTA, which make the mathematical framework of CIL rather dirty.
• Own implementation is more sustainable while integrated with Savu - I know where things are and can find and modify them quickly.
• Less dependence inside Savu to avoid that something is out of sync with CIL packaging.

Having said that, I still think having CIL as a plugin in Savu would be a good thing, especially to access other algorithms that require more effort than FISTA.

[dkazanc]

I even think that using FISTA from TomoPhantom is a bit too much due to additional dependence. I'd view Savu as a collection of independent self-contained plugins, which developers/users can modify if need to. If the method/algorithm is not directly exposed in Savu it is difficult to maintain it.