Sensitivity to scale

[sbysiak]

Thank you for the package!
The vivo score seems to be sensitive to scale of the features due to computing just the area under partial dependence plot while the x-axis may vary by orders of magnitude for various features.
I think it could be first standarized (at least as an option), otherwise the features with large difference between min and max value are always preferred, as can be seen e.g. in http://docs.mlinpl.org/conference/2019/posters/AnnaKozak.pdf

[kozaka93]

Thank you for the issue. In the first version, it was like that. Now we use weights based on empirical density or we average. Please check the density parameter in the latest version 0.2.1 available on CRAN and GitHub.