normalize.py 文件源码

def forward(self, x):
        size = x.size()
        x = x.view(x.size(0), -1)
        x = (x - th.mean(x, 1).unsqueeze(1)) / th.sqrt(th.var(x, 1).unsqueeze(1) + self.epsilon)
        if self.learnable:
            x =  self.alpha.expand_as(x) * x + self.beta.expand_as(x)
        return x.view(size)