STIR-python OSMAPOSL priors are GeneralisedPrior objects rather than child class priors

[robbietuk]

OSMAPOSL reconstruction object is defined and setup in python as self.recon. This setup process has a penalty defined (e.g. QP). The call self.recon.get_objective_function().get_prior() returns
<Swig Object of type 'stir::shared_ptr< stir::GeneralisedPrior< stir::DiscretisedDensity< 3,float > > > *' at 0x125b172f0>. This GeneralisedPrior object does not contain any methods specific to the QP/RDP and therefore evaluations of things like parabolic_surrogate_curvature() (QP), or get_gamma()/set_gamma() (RDP) cannot be performed.

Can get_prior() return the child class prior?, e.g. QP or RDP.

[KrisThielemans]

see #662 and #284

feel like trying what's in #284 (comment)?

[robbietuk]

I think I will. Unfortunately I don't think it can be put off for much longer.

[KrisThielemans]

could be very easy, but we'll only find out by trying. The SWIG list is quite helpful though. Let me know

[robbietuk]

I shall do. Let me learn SWIG first 😄

[KrisThielemans]

That could easily take up the rest if you PhD!

(edit src/swig/stir.i, add a single %factory line, see what happens...)

[robbietuk]

Is there any way we could do something similar to

STIR/src/swig/stir.i

Lines 1620 to 1626 in d1cd05c

	%inline %{
	template <class T>
	stir::PoissonLogLikelihoodWithLinearModelForMeanAndProjData<T> *
	ToPoissonLogLikelihoodWithLinearModelForMeanAndProjData(stir::GeneralisedObjectiveFunction<T> *b) {
	return dynamic_cast<stir::PoissonLogLikelihoodWithLinearModelForMeanAndProjData<T>*>(b);
	}
	%}

The python call would be something like stir.ToQP(self.recon.get_objective_function().get_prior()), which would return class type QP, with all the methods avalible.

I am guessing this would require an individual stir method for each prior (e.g. stir.ToRDP() and stir.ToPLS())and is not ideal for code clarity.

[KrisThielemans]

this is easy enough. We probably only need it for PriorWithParabolicSurrogate at the moment?

[robbietuk]

Ill give it a go for PriorWithParabolicSurrogate if not, Ill just do each prior individually 😄 I dont think think would be a perminent fix though. If it works, should we PR it to STIR master?

[KrisThielemans]

for what we need, this should work.

Regarding a PR, depends if we can get the SWIG people to help out first or not!

[robbietuk]

See robbietuk#10