search

bdwilliamson / vimp

Perform inference on algorithm-agnostic variable importance
http://bdwilliamson.github.io/vimp/
Other
21 stars 8 forks source link

Allow for pre-computed IPW weights #15

Closed bdwilliamson closed 2 years ago

bdwilliamson commented 3 years ago

This adds functionality requested in #14, where population weights (or pre-estimated weights) are available. Involves adding a new argument to vim and cv_vim and updating calls to, e.g., est_predictiveness_cv.

bdwilliamson commented 2 years ago

After thinking about this more, even if C = 1 for all observations, there projection of the full-data EIF onto the fully-observed variables is still used in the EIF (see Section 3.4 here). So there is, unfortunately, no way around doing this estimation in coarsened-data settings.