Algorithm class initial - to copy or not to copy?

[MargaretDuff]

In the algorithm class initial is usually copied internally in the algorithm so it is left unmodified e.g. in FISTA and ISTA:

CIL/Wrappers/Python/cil/optimisation/algorithms/FISTA.py

Lines 147 to 148 in 98860db

	self.x_old = initial.copy()
	self.x = initial.copy()

This is memory expensive - it means this memory can't be reused - however, it might make more sense to users for their initial values to be left unmodified.

Looking for any thoughts or opinions :)

[MargaretDuff]

Similarly, CGLS uses data only in the set_up and then, to keep data unchanged, does not reuse this memory buffer:

CIL/Wrappers/Python/cil/optimisation/algorithms/CGLS.py

Line 91 in 2097927

self.r = data - self.operator.direct(self.x)

Potentially by using this, a copy of the acquisition data could be saved

[purepani]

I would give the user the option to copy or reuse memory. Flexibility is good.

[MargaretDuff]

Thanks for your thoughts!