[DOC] Very high dim

[bthirion]

One setp toward #306 : populating the high dimension documentation module.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.28%. Comparing base (d84ea41) to head (433943c).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #425   +/-   ##
=======================================
  Coverage   99.28%   99.28%           
=======================================
  Files          23       23           
  Lines        1253     1253           
=======================================
  Hits         1244     1244           
  Misses          9        9           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jpaillard]

Would it make sense to add a figure from an example, like the fMRI or the 2D simulation?

[jpaillard]

This code is not executed in the rendered doc. Is there a way to run the code snippets?

[lionelkusch]

For the user guide, I don't think that it's a good idea to have code.
In my opinion, it's better to have an example associated with it and make reference to it.

[jpaillard]

Should we add a bit more details explaining that the clustering is constrained by pixel's neighborhood?

[jpaillard]

Suggested change

As hinted in :footcite:t:`meinshausen2009pvalues` an efficient way to deal with such configuration is to take the per-group average of the features: this leads to a *reduced design*. After inference, all the feature in a given group obtain the p-value of the group representative. When the inference engine is Desparsified Lasso, the resulting method is called Clustered Desparsified lasso, or **CluDL**.

As hinted in :footcite:t:`meinshausen2009pvalues` an efficient way to deal with such configuration is to take the per-group average of the features: this leads to a *reduced design*. After inference, all the feature in a given group obtain the p-value of the group representative. When the inference engine is Desparsified Lasso, the resulting method is called Clustered Desparsified lasso, or :py:func:`CluDL <hidimstat.clustered_inference>`.

I think this links to the API

[lionelkusch]

There are other models to cite it in this section; it should be interesting to add the link, for example Desparsified Lasso and EnCluDL.

[lionelkusch]

Can you try to stick to the maximum length of 80 characters?