Reference is now made to FIG. 7, showing a flowchart schematically representing an optional flow of operations 700 for computing a loss score, according to some embodiments of the present invention. In such embodiments, processor 590 computes in 710 for at least one of the plurality of set-operator prediction models a target set of features by applying the set operator of the at least one set-operator prediction model to the plurality of input sets of labels. Optionally, for each of the plurality of set-operator prediction models the respective target set of labels is the model target set of labels of the respective model. In 720, processor 590 optionally computes for the at least one set-operator prediction model a model set of labels by providing the at least one set operator prediction model's model group of features to one or more multi-label classification model 520. Optionally, a plurality of model set of labels each computed for one of the plurality of set-operator prediction models is the plurality output sets of labels computed in 608. In 730, processor 590 optionally computes for the at least one set-operator prediction model a difference between the model target set of labels of the at least one set-operator prediction model and the model set of labels of the at least one set-operator prediction model. Optionally, processor 590 computes another difference between the plurality of model sets of labels and the plurality of model target sets of labels. To compute the other difference, system 500 optionally implements the following method.