Any thoughts on omitting the preprocessing step in which NaNs are set to zero? Choosing how to handle missing data is a non-trivial modeling choice and having it happen automatically might cause problems for the analyst. My preference would be to drop any row with missings, but throwing an error is also a fine way to go. Thanks!
Any thoughts on omitting the preprocessing step in which NaNs are set to zero? Choosing how to handle missing data is a non-trivial modeling choice and having it happen automatically might cause problems for the analyst. My preference would be to drop any row with missings, but throwing an error is also a fine way to go. Thanks!