Distance-based subset selection revisited