Description

It would be useful if the current index of a Sampler instance could be accessed multiple times within an iteration without changing. For example, when both an SGFunction and a Preconditioner depend on the choice of subset and need to reference the same index.

Right now, I think that this can be done with:

current_index = algorithm.f.data_passes_indices[-1]

as long as the random samplers have a fixed seed. However, this feels like an unnecessary workaround rather than accessing the sampler itself

Possible Solutions

• Adding a current_index property to Sampler
  This would store the last sampled index and allow multiple accesses within the same iteration.

• Modifying Sampler.next() to accept algorithm.iteration
  This would ensure the index remains consistent when accessed multiple times within the same iteration but, I guess, would require some changes to the way stochastic functions work.

Either approach would make it easier to ensure consistency when using stochastic samplers.

Cheers

Sam

Environment

24.2.1.dev11+gab65d759.d20241129 gab65d759 3.10.12 (main, Jan 17 2025, 14:35:34) [GCC 11.4.0] linux

edit: a simpler way to get the current index, typo