Maximum entropy versus minimum risk and applications to some classical discrete distributions