Spaces:
Running
How about tasks of learning to rank and multi-task learning?
Any plans to add ranking and multi-task (classification & regression) tasks?
Heyho @Ouz-G , great question!
Currently, we have not yet started working on supporting either ranking or multi-task tasks.
But we plan to work on it in the (far) future.
Do you have any pointers for related work, datasets, or state-of-the-art (SOTA) methods for ranking or multi-task tasks?
This could help us a lot to bootstrap TabArena-Ranking or TabArena-Multi-Task!
From our current list of priorities, we would likely first tackle multi-task before ranking.
This discussion and comment might give some more context to what I think would be useful for multi-task stuff: https://github.com/dholzmueller/pytabkit/issues/17#issuecomment-2997123036
I think it would make sense to synthetically generate all sorts of data where multi-task loss functions are also randomly generated (e.g. 0.3 x MSE on subset A + 0.5 x MAE on subset B and 0.2 x binary cross entropy on subset C)