You could try dumping your dataframes to an in-memory SQLite instance using the to_sql method and then running queries on that. Not sure how performant the to_s

You could try dumping your dataframes to an in-memory SQLite instance using the to_sql method and then running queries on that. Not sure how performant the to_sql bit would be, but I'd expect joins in SQLite to be blazing fast compared to Panda's joins.