Translations:Loss Functions/33/en

An important consideration is whether the loss is calibrated — i.e., whether minimising it yields well-calibrated predicted probabilities. Cross-entropy is a proper scoring rule and produces calibrated probabilities, while hinge loss does not.