查看网络grad

输出网络结构：

print("auxi=",nn.Sequential(*list(self.auxi_network.children())))

对每一层的访问方式。模型打印出来前面带标号的，如(0)，可以直接用下标[0]访问。如果前面是带名字的，如(con1)，则需要用属性名.conv1访问。

如：

print("auxi=",nn.Sequential(*list(self.auxi_network.children()))[0][4][1].conv1.weight.grad)

打印模型权重和梯度：

net = Net()
print(net.conv21.bias)
print(net.conv21.bias.grad)
print(net.conv21.weight)
print(net.conv21.weight.grad)

打印Sequential序列中的值：

for i,m in enumerate(net.conv1.children()):
    if isinstance(m, nn.Conv2d):
        print("net.conv1."+str(i)+"(Conv2d).weight = ",m.weight)
        print("net.conv1."+str(i)+"(Conv2d).weight.grad = ",m.weight.grad)
    elif isinstance(m, nn.BatchNorm2d):
        print("net.conv1."+str(i)+"(BatchNorm2d).weight = ",m.weight)
        print("net.conv1."+str(i)+"(BatchNorm2d).weight.grad = ",m.weight.grad)