‘CrossEntropyLoss‘ object has no attribute ‘backward‘报错解决

CrossEntropyLoss没有backward属性

错误代码如下，调用封装函数直接会返回CrossEntropyLoss函数相当于直接调用了CrossEntropyLoss的backward,所以报错，应该对CrossEntropyLoss的返回值使用backward

def CrossEntropyLoss_func(output, target):
    return nn.CrossEntropyLoss()

loss = CrossEntropyLoss_func(output, target)
loss.backward()

正确代码如下，

def CrossEntropyLoss_func():
    return nn.CrossEntropyLoss()

my_loss = CrossEntropyLoss_func()
loss = my_loss(output,target)
loss.backward()

#也就是
my_loss = nn.CrossEntropyLoss()
loss = my_loss(output,target)
loss.backward()

版权声明：本文为L888666Q原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接和本声明。