Clustering in advanced data analytics Bottom up? Top down? Or both?