outcome of our analysis is that, for classiﬁcation, the hinge loss appears
to be the loss of choice. Other things being equal, the hinge loss leads to
a convergence rate practically indistinguishable from the logistic loss rate
and much better than the square loss rate. Furthermore, if the hypothesis
space is suﬃciently rich, the bounds obtained for the hinge loss are not
loosened by the thresholding stage.