currently, using any ml model results in loading keras, tensorflow and sklearn. Splitting up files should make imports better.