CCPi Regularisation toolkit plugin: bugfix, docstrings, unittests, remove unused functions

[paskino]

closes #931

[paskino]

Please @epapoutsellis and/or @jakobsj can you advise on what to do with LLT_ROF? There are 2 regularisation parameters, and I currently have multiplied both of them.

[epapoutsellis]

I think we delayed a lot adding tests for the regularisation toolkit. The only option easy and fast is via cvxpy.

[paskino]

I don't know. Tests should be implemented in the CCPi Regularisation package. Some are already implemented, though they test each other or the C/OpenMP vs the CUDA code.

But the tests you want to implement would be good to have anyway. Open an issue about it.

[paskino]

An issue is already open for TNV: TomographicImaging/CCPi-Regularisation-Toolkit#146

[paskino]

Actually we only use FGP_TV, FGP_dTV, TGV and TNV. So we should scrap all the other plugins and add appropriate unittests for these in the CCPi-Regularisation toolkit

Also check if there are unittests for the plugin functionality.

[paskino]

OK, I removed all the regularisng functions we do not use. I added unit tests to compare our plugin via proximal and the direct call of the regulariser.

Could you please confirm the defaults I chose?

[paskino]

I also added __rmul__ to this PR, so that constructs like

f = alpha * FGP_TV()

return a FGP_TV function with an internal regularisation parameter multiplied by alpha rather than a ScaledFunction

[paskino]

Jenkin's happy.

[gfardell]

I think we need docstrings for the 4 we're wrapping, but that's not really the fault of this PR!

[epapoutsellis]

The default ratio for the TGV parameters should be alpha0\alpha1 \in [1,2]

https://github.com/vais-ral/CCPi-Regularisation-Toolkit/blob/413c6001003c6f1272aeb43152654baaf0c8a423/demos/demo_cpu_regularisers.py#L312

[paskino]

Notice that I removed the printing parameter from the calls as useless. I changed to the max_iterations member in each function as that we a more appropriate name.

[paskino]

Jenkin's happy

[epapoutsellis]

LGTM happy to merge