vimp: vimp: Perform Inference on Algorithm-Agnostic Intrinsic...

vimpR Documentation

vimp: Perform Inference on Algorithm-Agnostic Intrinsic Variable Importance


A unified framework for valid statistical inference on algorithm-agnostic measures of intrinsic variable importance. You provide the data, a method for estimating the conditional mean of the outcome given the covariates, choose a variable importance measure, and specify variable(s) of interest; 'vimp' takes care of the rest.


Maintainer: Brian Williamson Contributor: Jean Feng

Methodology authors:

  • Brian D. Williamson

  • Jean Feng

  • Peter B. Gilbert

  • Noah R. Simon

  • Marco Carone

See Also


Other useful links:


The packages that we import either make the internal code nice (dplyr, magrittr, tibble, rlang, MASS, data.table), are directly relevant to estimating the conditional mean (SuperLearner) or predictiveness measures (ROCR), or are necessary for hypothesis testing (stats) or confidence intervals (boot, only for bootstrap intervals).

We suggest several other packages: xgboost, ranger, gam, glmnet, polspline, and quadprog allow a flexible library of candidate learners in the Super Learner; ggplot2 and cowplot help with plotting variable importance estimates; testthat and covr help with unit tests; and knitr, rmarkdown, and tidyselect help with the vignettes and examples.

bdwilliamson/npvi documentation built on Feb. 13, 2023, 9:58 a.m.