This message was deleted.

There's a

dropna

argument in preprocess and fit that defaults to True, you can set

dropna=False

to keep those rows

@José Morales thanks, I wonder if default = False would have been a better option, just run the version with dropped NaNs and adjusted R2 went all the way to negative in global linear model.

Not all models can handle nulls, that's why the default is True