Nilearn non_parametric_inference


I’ve noticed that if I rerun non_parametric_inference function from nilearn I get slightly different results.
Are there any other ways to ensure reproducibility apart from setting a random_seed?
I’m just wondering what can theoretically prevent you from simply selecting the seed that you like most.


The only good answer I know to that is to increase the number of permutations to tame inference variability.
Seed hacking would be a deleterious indeed.