In this data format the input dataset is given in two files: One file with a (dense) representation of features and one file with the (sparse) labels.