損失函數

下表概述了MLlib支持的損失函數及其梯度和子梯度：

	損失函數$L(w;x,y)$	梯度或子梯度
hinge loss	$max\{0, 1-yw^Tx\}$	$\left\{ \begin{array}{ll} -yx ~~~ 若~yw^Tx \lt1, \\ 0 ~~~~~~~~~otherwise \end{array} \right.$
logistic loss	$log(1+exp(-yw^Tx)),~y \in \{-1,+1\}$	$-y(1-\frac{1}{1+exp(-yw^Tx)})x$
squared loss	$\frac{1}{2}(x^Tx-y)^2,~y \in R$	$(w^Tx-y)x$