CIL icon indicating copy to clipboard operation
CIL copied to clipboard
verified publisher icon TomographicImaging

SPDHG optimal sampling

Open MargaretDuff opened this issue 2 years ago • 1 comments

https://github.com/TomographicImaging/CIL/blob/a2e45fe0aafe6329ec9027bda1cf262d4aff0284/Wrappers/Python/cil/optimisation/algorithms/SPDHG.py#L50C44-L50C44

Instead of initialising the probabilities for random sampling in SPDHG to be uniform, we could also add in optimal sampling example 5 in Matthias SIAM optimisation paper

MargaretDuff avatar Aug 15 '23 09:08 MargaretDuff

I think in the paper they show optimal sampling when they know their functions are strongly convex. We don't yet have strong convexity options in SPDHG, although maybe we should!

MargaretDuff avatar Aug 22 '24 11:08 MargaretDuff