5-用PyTorch实现线性回归

下面是损失函数

下面是优化器
下面通过model.parameters()可以获得model中所有的参数

点击查看代码

import torch
from torch import device

x_data = torch.tensor([[1.0], [2.0], [3.0]])
y_data = torch.tensor([[2.0], [4.0], [6.0]])

class LinearModel(torch.nn.Module):
    def __init__(self):
        super(LinearModel, self).__init__()
        self.linear = torch.nn.Linear(1, 1) # 权重和偏置

    def forward(self, x):
        y_pred = self.linear(x)
        return y_pred

model = LinearModel() # 定义模型

# 定义损失函数
criterion = torch.nn.MSELoss()
# 定义优化器
optimizer = torch.optim.SGD(model.parameters(), lr=0.01) # 优化器对哪些参数进行更新

for epoch in range(5000):
    y_pred = model(x_data)
    loss = criterion(y_pred, y_data) # 计算损失
    print(epoch, loss.item())

    optimizer.zero_grad() # 梯度清零
    loss.backward() # 反向传播
    optimizer.step() # 参数更新

print('w=', model.linear.weight.item())
print('b=', model.linear.bias.item())

x_test = torch.tensor([4.0])
print('predict(4)=', model(x_test).item())