PyTorch深度学习实践-刘二大人-反向传播作业

 

 

import matplotlib.pyplot as plt
import torch

# y = w1*x2 + w2 * x + b,注意超参数学习率的设置，这里设置为0.01
x_data = [1, 2, 3]
y_data = [2, 4, 6]

loss_list = []
w1 = torch.Tensor([1])
w2 = torch.Tensor([1])
b = torch.Tensor([1])

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


def forward(x):
    return w1*x**2+w2*x+b        #W是tensor，tensor的运算后是建立计算图


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


for epoch in range(100):
    for x, y in zip(x_data, y_data):
        l = loss(x, y)      #建立计算图
        l.backward()        #每次调用backward后，计算图消失
        print('\tgrad:', x, y, w1.grad.item(), w2.grad.item(), b.grad.item())
        w1.data = w1.data - 0.01 * w1.grad.data  #W是tensor,.data属性和W.grad.data也是tensor，但其不建立计算图
        w2.data = w2.data - 0.01 * w2.grad.data
        b.data = b.data - 0.01 * b.grad.data

        w1.grad.data.zero_()     #防止w的梯度累加起来,因此需要清空
        w2.grad.data.zero_()
        b.grad.data.zero_()

    loss_list.append(l.item()) #item()将tensor转换成标量
    print('progress:', epoch, l.item())
print("predict (after training)", 4, forward(4).item())

plt.plot(loss_list)
plt.ylabel('loss')
plt.show()