Modeling long-term dependencies and short-term correlations in patient journey data with temporal attention networks for health prediction.