Thanks for sharing this package and writing the paper.
It would be nice if this package also supported binary multi-label classification problems.
What would be the best way to aggregate for instance "Smallest Margin" computed for each label into a per-sample score in your opinion?