Turn FeatureExtractor into an abstract class that handles the if(Labeled) casting issues in all FeatureExtractors
FeatureExtractor Interface: Type (maybe Datum maybe not) → Datum, Readers output a Type (for now we can stick with Collection of datums)
Models take Datums (exceptions may exist like NetflixMovieModel but not in code base)
pipelines?
what is the role of vectors in the feature pipeline?