pytorch实现线性回归

import torch
import matplotlib.pyplot as plt

## prepare dataset
x_data = torch.Tensor([[1.0], [2.0], [3.0]])
y_data = torch.Tensor([[2.0], [4.0], [6.0]])

## design model using class
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()

## construct loss and optimizer
criterion = torch.nn.MSELoss(size_average=False)
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

## training cycle
losses = []
for epoch in range(1000):
    y_pred = model(x_data)
    loss = criterion(y_pred, y_data)
    losses.append(loss)
    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)

l = [i for i in range(1000)]
plt.plot(l, losses)
plt.xlabel('epoch')
plt.ylabel('loss')
plt.show()