Some models are able to produce probabilistic output that follows some distribution.
This would mean you can model uncertainty much better than simply based on entropy / other heuristics.
Therefore:
- it would be nice to support distributional output
- it would also be nice to support the function to produce model output as a parameter