pytorch 一些 Loss Function

nn.L1Loss()

loss(x,y) = 1/n ∑ |xi - yi| n是元素的总个数

shape:

input: (N,) 是任何额外的维度

target: (N,*) same as input

output: scalar ，标量

nnMSELoss()

loss(x,y) =1/n ∑（xi-yi）²

shape:

input: (N,*)

target: (N,*)

output: scalar

nn.CrossEntropyLoss()

单目标二分类或者多分类

loss = -log[ exp(x[class])/ (∑ exp( x[j]))]

shape:

input: (N,C) C = number of classes

target : (N)

NN.BCELoss()

单目标二分类交叉熵函数,Binary_Corss_Entropy

loss = -1/n ∑{ t[i]*log(o[i])-(1-t[i]) *log(1-o[i])} i = 0,1

shape:

input: (N,*)

output: (N,*)