originally mentioned in #197. ideally this will smoothly create multi-indexes as/when necessary (#164), just like the pandas method