How do we encode multi-label or sparse task labels?

[mmcdermott]

In our label schema, if a task is multi-label (e.g., predict all the diagnoses a patient will receive in this stay), how do we want to encode that? What if a task is sparse and multi-label (e.g., predict the max lab value that will be seen for all labs measured in this stay)?

I don't think this is urgent at all, but it is worth contemplating to either capture in documentation or in future versions (or to have a place for catching discussions about it)