I just realized that Hidimstat has no narrative documentation. It is clearly the main missing piece atm. This something we need to work on.
It has to be tightly coupled with examples as Nilearn and sklearn. Adding stuff on pdp, marginal vim etc. will be important in that respect.
It might be organized as follows:

• Definition of variable importance
  ** variable importance, variable selection
  ** conditional vs marginal variable importance
  ** roadblocks toward variable importance computation: dimensionality and correlation
  ** statistical inference / confidence intervals
• A review of vi methods, organized according to the target criterions
• Very high dimensional case (EncludL etc)