1

I am doing regression modeling, where I have a continuous value I want to predict using many (~60) features.

However, many of the samples in my dataset contain NaN features. Some features are very reliable, but other features may be significantly or mostly NaN for all samples.

I want to use the reliable features I do have to impute the missing values of the non-reliable ones. I don't want to simply replace the missing values naively with some number (like the mean, or median for instance) like scikit impute offers.

But no package I can find in python (except maybe fancyimpute) will do this. Yet there are many packages I can find in R that will do this.

Should I send my data to R, or are there other python packages I don't know about for smart imputation? In general, what are some good environments and packages that could make it easy to start doing this? Where should I be directing my efforts.

1
  • You should do your imputation in R. Python's statistical development is probably about a decade behind R. It has apparently mastered most of the plotting functionality and much of the regression analysis but the "real" stats research and development is still in R. Python is ahead of R in data management and access to big data engines.
    – 42-
    Jun 28, 2018 at 17:52

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.