[DOC] Deleted example: Support recovery on MEG data

[jpaillard]

Thanks to the version dropdown menu in the doc, I realized that in version 0.1.0, we had an example, "Support recovery on MEG data"
Do you remember why we dropped it? Should we add it back?
@bthirion @lionelkusch

[bthirion]

It was adding quite a few dependencies, and some additional complexity (spatial/temporal structure).

[jpaillard]

Would you say that adding a dependency on mne-python for the doc is not an option because to heavy?
If so we can close this. Otherwise we can think about how to integrate M/EEG examples with limited complexity.

[bthirion]

This is truly an open question.
My view is that indeed having to install MNE is too much for hidimstat. I'm worried of doc no longer building, timeouts etc. Yet we have a dependency on Nilearn, so it makes sense to consider both cases together.

[jpaillard]

The list of dependencies for MNE does not seem unreasonable to me (see below) and would only be needed for the doc.

Fetching a dataset would increase the time to build the example, but if it remains tractable for fMRI, it can probably also be managed for M/EEG.

Should I try to open a PR with an example to see how long it takes to build the doc? If yes, would you have a suggestion?

Python ≥ 3.10
NumPy ≥ 1.26
SciPy ≥ 1.11
Matplotlib ≥ 3.8
Pooch ≥ 1.5
tqdm
Jinja2
decorator
lazy-loader ≥ 0.3
packaging

[bthirion]

To me this is low priority.
If you wna tto give a try, just recreate the example that was set initially by @Jache