FeatureRequest: Anonymous (dataset) publication

[PGijsbers]

Current: All entities are publicly linked to a user. There is no way to hide which user uploaded the entity.

Request: Allow entities (datasets, flows, tasks, runs, ...) to be created without publicly showing their creator. Optionally making the relationship public at a later date.

Use case: Anonymously sharing datasets and/or experimental results for a paper under review.

Impacted Projects: This impacts all parts of OpenML. It requires additional data stored for each entity in the back-end and the front-end and connector packages to allow a user field on an entity being null/absent (depending on implementation).

Suggested Implementation: One way to do this with minimal overhead might be to introduce a second column in each entity, we keep the foreign key to the user but also add an extra column visible_user which indicates which user is publicly associated with the entity (for now, it'd be either the same user or null). The data sent to the front-end/connector would be of visible_user only and it gets processed in the same way as the user id currently is (except it should allow for null). Initially, I would limit this to datasets and only anonymous to public status and expand from there.

[PGijsbers]

Discussed at virtual hackathon 17-11-2022.
Present: @joaquinvanschoren, @prabhant, @PGijsbers
Conclusion: Accepted.

Instead of using null a token user should be made for anonymous uploads. This makes the reason for its absence explicit and would also be able to extend to deleted users and other scenarios in the future.