The reliability of Marlatt's taxonomy for classifying relapses.