Ok, here's one example: Want to know whether training/test data sets have the same distribution? Try to predict if a point is train/test.