The loss function assigning to a predicted probability distribution and observed outcome . It is the negative logarithmic scoring rule; its expected value under the true probability distribution is cross-entropy. Predicting zero probability for an observed outcome gives infinite loss.
New to topics? Read the docs here!