clj-djl.training.loss
hinge
(hinge)(hinge name)(hinge name margin weight)The hinge loss is used for maximum-margin classification.
l1
(l1)(l1 name)(l1 name weight)Least absolute deviations loss function minimizes the absolute differences between the estimated values and the existing target values.
l2
(l2)(l2 name)(l2 name weight)Least square errors loss function minimizes the squared differences between the estimated and existing target values.
masked-softmax-cross-entropy
(masked-softmax-cross-entropy)(masked-softmax-cross-entropy name)(masked-softmax-cross-entropy name weight class-axis sparse-label from-logit)sigmoid-binary-cross-entropy
(sigmoid-binary-cross-entropy)(sigmoid-binary-cross-entropy name)(sigmoid-binary-cross-entropy name weight from-sigmoid)sotfmax-cross-entropy
(sotfmax-cross-entropy)(sotfmax-cross-entropy name)(sotfmax-cross-entropy name weight class-axis sparse-label from-logit)