Stratified cross-validation for multi-label data sets

Greedily assign samples into folds based on their labels

The algorithm partitions a binary multi-label Y into N folds. The returned vector assigns samples into the folds such way, that the same distribution of positive and negative samples of each label is approximately maintained across all folds.

The implementation can be used as a drop-in replacement for CVPARTITION.

Reference: On the Stratification of Multi-label Data by Sechidis, Konstantinos & Tsoumakas, Grigorios & Vlahavas, Ioannis.