Leveraging Importance Weights in Subset Selection