# 来自B站刘二大人
import torch
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)
    def forward(self, x): 
        y_pred = self.linear(x) 
        return y_pred
    
model = LinearModel()
criterion = torch.nn.MSELoss(size_average=False)
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

for epoch in range(1000):
    y_pred = model(x_data)
    loss = criterion(y_pred, y_data) 
    print(epoch, loss.item())
    
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()
    
print('w = ', model.linear.weight.item())
print('b = ', model.linear.bias.item())
x_test = torch.Tensor([[4.0]]) 
y_test = model(x_test) 
print('y_pred = ', y_test.data)