Why not analyzing a dataset as if it were an image (with brushes, erasers,)?