Analysis: Don't do big-data science backwards