Any chance for ORM/pandas-like interface to augment the sql-like interface?

[FishmanL]

eg supporting filter/groupby methods on the returned data object
should just be a matter of converting to AST under the hood?

[joshua-oss]

Hmm, good question. The way the SQL layer works is to:

1. Intercept a query request that is formatted as a string
2. Parse to AST and validate the SQL to ensure it's a query that can be safely done in a private way
3. Rewrite the query in the AST to ensure privacy. For example, add reservoir sampling or bounds clipping
4. Convert the rewritten query to a string
5. Execute the rewritten query string against the database connection

Steps 2-4 need to be handled by SmartNoise to make sure we don't violate privacy. But it should be straightforward to feed step 1 with the output from an ORM-style query. As long as the ORM layer doesn't generate queries which are inconsistent with privacy, it will work. I believe SQLAlchemy allows you to grab a SQL query string from the ORM query, so it should "just work". I will try to whip up a sample later next week. I don't know if it's possible to do that from Pandas, though. If there is a way from Pandas to grab the SQL generated from the filter/groupby methods, it would presumably work.

[FishmanL]

Something like https://github.com/AmirPupko/pandas-to-sql looks helpful here!