《PyTorch深度学习实践》刘二大人 第4讲反向传播

（刚开始学这方面知识，如有错误，请大家指正）

一、课堂代码+运行结果

import torch

x_data = [1.0, 2.0, 3.0]
y_data = [2.0, 4.0, 6.0]

w = torch.tensor([1.0])
w.requires_grad = True


def forward(x):
    return w * x


def loss(x, y):
    y_pred = forward(x)
    return (y_pred - y) ** 2


print(f"predict (before training), {4, forward(4).item()}")

for epoch in range(100):
    for x, y in zip(x_data, y_data):
        l = loss(x, y)
        l.backward()
        print("\tgrad: ", x, y, w.grad.item())
        w.data = w.data - 0.01 * w.grad.data
        
        w.grad.data.zero_()

    print(f"progress: {epoch, l.item()}")

print(f"predict (after training), {4, forward(4).item()}")

二、课后作业

模型：

损失函数：

训练集：输入=[1, 2, 3]，输出=[2, 4, 6]

测试集：输入=4，猜测输出应该是8

利用计算图，算出    

import torch

x_data = [1.0, 2.0, 3.0]
y_data = [2.0, 4.0, 6.0]

w1 = torch.tensor([1.0])
w2 = torch.tensor([1.0])
b = torch.tensor([1.0])

w1.requires_grad = True
w2.requires_grad = True
b.requires_grad = True


# 前向传播
def forward(x):
    return w1 * x * x + w2 * x + b


# 损失
def loss(x, y):
    y_pred = forward(x)
    return (y_pred - y) ** 2


print(f"predict (before training), {4, forward(4).item()}")
# 可以改变range 100 1000 10000 100000 对比效果
for epoch in range(10000):
    for x, y in zip(x_data, y_data):
        l = loss(x, y)
        # 反向传播，求梯度
        l.backward()

        # 利用梯度更新w1, w2, b
        w1.data -= 0.01 * w1.grad.data
        w2.data -= 0.01 * w2.grad.data
        b.data -= 0.01 * b.grad.data
        # print(f"progress: w1 = {w1.data}, w2 = {w2.data}, b = {b.data}")

        # 将w1, w2, b的梯度置为0
        w1.grad.data.zero_()
        w2.grad.data.zero_()
        b.grad.data.zero_()

print(f"predict (after training), {4, forward(4).item()}")  # 理想是 4, 8

训练次数为100时：

训练次数为1000时：

训练次数为10000时：

训练次数为100000时：（预测与理想越来越接近）

计算图：

     (在 x = 1的情况下)