Use this model type when any number of the supplied labels can apply to each example (i.e. each label is independent from the others). This is equivalent to training a separate classifier to predict whether or not each possible label applies to a provided document. The probabilities returned by a call to `predict()` can each range from 0. to 1. and are not required to sum to 1. Examples of multi-label classification include predicting the topics an article covers or predicting what themes a user mentioned in a review.
In this case you can supply multiple labels per data point when adding data to your custom collection.
// Multi-label Collections are currently not supported by the Java client library.