* allow partial evaluation of dataframe before prediction * resume spark test * comments * Run unit tests after building JVM packages