In many settings it's useful to treat missingness as a new category when it affects a categorical column. It would be great to have some transformer (with a meaningful name) that transforms missing values in each column of a given table into a new category (level) for that column.
This would make it possible to use encoders or machine learning models that don't have a default way to deal with missingness and it's a popular strategy to treat missingness meaningfully instead of using imputation.
Happy to provide guidance to a new contributor. The API for implementing a new model is documented here. You can find other transformer implementations here.
ablaom
commented
3 months ago
Motivation
In many settings it's useful to treat missingness as a new category when it affects a categorical column. It would be great to have some transformer (with a meaningful name) that transforms missing values in each column of a given table into a new category (level) for that column.
This would make it possible to use encoders or machine learning models that don't have a default way to deal with missingness and it's a popular strategy to treat missingness meaningfully instead of using imputation.