Proposition for new feature : Add option to compute grouped variable importance

[ptaconet]

It would be very nice to have an option to compute permutation variable importance for groups of variables (instead of single variables), i.e. to have an insight on how selected variables altogether influence the prediction, as it was for instance proposed in this package : https://rdrr.io/cran/RFgroove/man/varImpGroup.html based on this paper : https://hal.inria.fr/hal-01084301v2/document

thanks for your very useful package :)

[bgreenwell]

Thanks for the suggestion @ptaconet, we’ve wondered about this for a while now: #62. I’ll look into it!

[ptaconet]

I had missed that one ... ! thanks a lot :)

[bgreenwell]

Wonder if this could also be useful in quantifying the importance of interaction effects?!

[ptaconet]

Not sure I am expert enough to elaborate on this... :S there seem to have tools existing to quantify the importance of interaction effects, e.g. Friedman's H-statistic