Jérôme Dockès

> yeah but for one of those I was getting an "empty file" error: will try to investigate a bit more there is some documentation in https://github.com/nilearn/nilearn/blob/main/nilearn/datasets/tests/_testing.py but it probably...

@Remi-Gau as you are looking at cleaning up some tests see also this one

from numpy doc: > This is a convenience, legacy function that exists to support older code that uses the singleton RandomState. Best practice is to use a dedicated Generator instance...

seeding the global RandomState in a function is a bad idea in general because it modifies a global object -- other functions that use random numbers will also start getting...

that would work, with a small change: `RandomState` is the legacy API. now you would use `np.random.default_rng(42)`. Also it doesn't achieve the effect @ymzayek 's suggestion did, ie forcing random...

`check_random_state` is the scikit-learn utility; it can take either `None` (meaning don't seed), a number (seed), or a RandomState (passthrough) and returns a random state > Also... Why 42? Is...

I think the slight asymmetry isn't a big deal for most users, so I'll close this as a "won't fix" for now; and there is always the option of removing...

> That's definitely a fair point. I'd be in favour of solving it in a different PR though, because it's not clear to me whether `_get_vertexcolor()` or `load_surf_data()` should do...

> Most methods modified by this PR now have `bg_map_rescale=True` by default (see `plot_surf()`, `plot_surf_stat_map()`, `plot_surf_roi()`) in order not to break current pieces of code like the following one: I...

> because the other missing labels will still be present in `masker.labels[maskers.labels_]` are you sure that is the case? AFAICT only non-empty regions in the mask get dimensions in the...