Open chadhazlett opened 7 years ago
What about in the returned K, U, etc?
I would propose we leave out the missingness in those objects and just keep NAs in all the fit/predict type objects. Main motive for this would be that if people are generating predictions by multiple procedures then it needs to be easy to weave them together.
Ideally we'd deal with missingness much as lm does -- i.e. it is na.omitted internally but then the predictions etc. are put back into the full length vector with the NA intermixed.