How about adding sampling methods that address class imbalance? The simplest idea would be to add a function to downsample all classes to the number of elements of the smallest class or oversampling all classes to the number of elements of the biggest class. Another option is some sort of weighted bootstrap, where probability of picking an element from a class is proportional to the class size -- that would probably need to include some sort of weighing parameter and depends on the sampling method.
If you think this is relevant I could fire a pull request with some ideas.
How about adding sampling methods that address class imbalance? The simplest idea would be to add a function to downsample all classes to the number of elements of the smallest class or oversampling all classes to the number of elements of the biggest class. Another option is some sort of weighted bootstrap, where probability of picking an element from a class is proportional to the class size -- that would probably need to include some sort of weighing parameter and depends on the sampling method.
If you think this is relevant I could fire a pull request with some ideas.