Jérôme Dockès

You are right, if we decide to push to pypi from the ci we will use an api token. but that still requires making the token available to the github...

note that's just one of the points in the issue description, I don't have a strong opinion re. the other ones. If there was a way to avoid having to...

can you try replacing collection_ids=[1952] with collection_id=1952?

I will open a PR for this (using fetch_neurovault and specifying collection_ids or image_ids to raise an error).

actually, I several things need to be improved in the Neurovault fetcher. maybe we can make a list here. @CherkaouiHamza

 here is a profiling of the script provided by @RaphaelMeudec ; not surprisingly most of the time is spent resampling and regressing the inputs on the atlas maps [profile.svg.zip](https://github.com/nilearn/nilearn/files/8453631/profile.svg.zip)

you are using coordinates on the inflated surface to sample the volume image. Use the true coordinates for the projection and then the inflated surface for plotting only: ```diff -glasser_rois_surf...

to help use this function we could add an Examples section to the docstring? also I think we can get rid of the warning that it is experimental, I don't...

I was thinking of a tiny example in the docstring itself (not an example in the gallery). something like ``` Examples -------- >>> from nilearn import datasets, surface >>> >>>...

Sorry for the late reply @magdalenasabat. Don't feel like a dummy! :D certainly the documentation can be improved. I think the confusion comes from the term "projection". The transformation is...