Enviar por SMS: Learning naive Bayes regression models with missing data using mixtures of truncated exponentials