I'm currently looking for a graphical lasso implementation allowing to specify different amount of regularization per variable.
As precised in the original publication (End of page 6, Remark ; also joint image), it can make sense to have a regularization that is different for each pair of features.
I know an R implementation allowing to pass the rho (penalizing parameter) as a float OR an array (features * features), but I couldn't find a matching one in R.
The implementation of graphical lasso in sklean doesn't seem to allow such an option since the penalizing parameter (alpha) should be a float.
Do you know a way to use this implementation tha way I would need to, or an other package containing such a flexible implementation ?
Thansk a lot for you help !
Rémi