MXNet 与 Pytorch 的 autograd

Pytorch 定义神经网络

import torch
from torch import nn
from torch.nn import functional as F


class Net(nn.Module):
    def __init__(self):
        super().__init__()
        # 1 个输出通道, 6 个输出通道, 5x5 的卷积核
        self.conv1 = nn.Conv2d(1, 6, 5)
        self.conv2 = nn.Conv2d(6, 16, 5)
        # 仿射运算: y = Wx + b
        self.fc1 = nn.Linear(16 * 5 * 5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x):
        # Max pooling over a (2, 2) window
        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
        # 对于方形仅需要给定一个参数
        x = F.max_pool2d(F.relu(self.conv2(x)), 2)
        x = x.view(-1, self.num_flat_features(x))
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x

    def num_flat_features(self, x):
        size = x.size()[1:]  # all dimensions except the batch dimension
        num_features = 1
        for s in size:
            num_features *= s
        return num_features


net = Net()
print(net)

Net(
  (conv1): Conv2d(1, 6, kernel_size=(5, 5), stride=(1, 1))
  (conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))
  (fc1): Linear(in_features=400, out_features=120, bias=True)
  (fc2): Linear(in_features=120, out_features=84, bias=True)
  (fc3): Linear(in_features=84, out_features=10, bias=True)
)

params = list(net.parameters())
print(len(params))
print(params[0].size())  # conv1 的 .weight

10
torch.Size([6, 1, 5, 5])

type(params[0])  # 查看数据类型

torch.nn.parameter.Parameter

x_input = torch.randn(1, 1, 32, 32)
out = net(x_input)
print(out)

tensor([[-0.0391, -0.0291, -0.0254, -0.0962,  0.1101,  0.0504,  0.0392,  0.0566,
          0.0468,  0.0377]], grad_fn=<AddmmBackward>)

Zero the gradient buffers of all parameters and backprops with random gradients（将具有随机梯度的所有参数和后向传播的梯度缓冲区归零）:

net.zero_grad()
out.backward(torch.randn(1, 10))

计算损失函数：

output = net(x_input)
target = torch.randn(10)  # a dummy target, for example
target = target.view(1, -1)  # make it the same shape as output
criterion = nn.MSELoss()

loss = criterion(output, target)
print(loss)

tensor(0.9708, grad_fn=<MseLossBackward>)

backprops（后向传播）

print(loss.grad_fn)  # MSELoss
print(loss.grad_fn.next_functions[0][0])  # Linear
print(loss.grad_fn.next_functions[0][0].next_functions[0][0])  # ReLU

<MseLossBackward object at 0x000002151D3BA908>
<AddmmBackward object at 0x000002151D3BA2B0>
<AccumulateGrad object at 0x000002151D3BAB00>

net.zero_grad()     # zeroes the gradient buffers of all parameters

print('conv1.bias.grad before backward')
print(net.conv1.bias.grad)

loss.backward()

print('conv1.bias.grad after backward')
print(net.conv1.bias.grad)

conv1.bias.grad before backward
tensor([0., 0., 0., 0., 0., 0.])
conv1.bias.grad after backward
tensor([ 0.0073,  0.0028, -0.0098,  0.0208, -0.0209, -0.0117])

手动更新权重：

learning_rate = 0.01
for f in net.parameters():
    f.data.sub_(f.grad.data * learning_rate)

f.grad

tensor([ 0.3875, -0.0547, -0.0328, -0.1468, -0.0062,  0.2169,  0.1387, -0.1924,
        -0.2262,  0.2408])

torch.optim 更新权重：

from torch import optim

# create your optimizer
optimizer = optim.SGD(net.parameters(), lr=0.01)

# in your training loop:
optimizer.zero_grad()   # zero the gradient buffers
output = net(x_input)
loss = criterion(output, target)
loss.backward()
optimizer.step()    # Does the update

注意：需要手动设置 optimizer.zero_grad() 在 Backprop 阶段出现的梯度累积，以避免 CUDA 显存不足。