Stochastic CIL - unexpected behaviour when the choice of number of subsets does not divide the number of projections

[MargaretDuff]

Description

Playing around with the new stochastic framework. In this 3D dataset with 900 projections we tried 20,30,35 and 50 subsets. We got odd results for 35 subsets, noting that this does not divide 900:

To do:

• [ ] Check this on another (smaller) dataset.
• [ ] Try playing with the probabilities in the sampler based on the number of projections in each subset

[MargaretDuff]

@epapoutsellis or @paskino - would appreciate any thoughts you have on this!

[paskino]

It looks to me that everyone is actually making sure that the number of subsets are divisors of the number of projections. It doesn't quite make sense to me, though.