Is it possible with neural forecast to add transformer for categorical variable and target transformer like we do with scikit-learn ?