基于pytorch的线性回归算法并比较了Adam与SGD的收敛速度

from torch import nn
from torch import optim as optimizer
import torch
from torch.autograd import Variable
import matplotlib.pyplot as plt
import numpy as np

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.linear = nn.Linear(1, 1)
    def forward(self, x):
        out = self.linear(x)
        return out

x = [[1.0], [1.3], [1.9], [2.8]]
y = [[0.9], [2.6], [2.1], [3.0]]

X = Variable(torch.Tensor([[1.0], [1.3], [1.9], [2.8]]))
y = Variable(torch.Tensor([[0.9], [2.6], [2.1], [3.0]]))

model_sgd = Model()
model_ad = Model()
loss_function = nn.MSELoss()

sgd = optimizer.SGD(model_sgd.parameters(), lr=0.01)
ad = optimizer.Adam(model_ad.parameters(), lr=0.01)

loss_sgd = []
for epoch in range(1000):
    out = model_sgd(X)
    sgd.zero_grad()
    loss = loss_function(out, y)
    loss.backward()
    sgd.step()

    print('epoch={}, loss={}'.format(epoch, loss.data.numpy()))
    loss_sgd.append(loss.data.numpy())

plt.scatter(x, y)

Y = model_sgd(torch.Tensor([[x] for x in np.linspace(1, 3, 100)])).data.numpy()
print(Y)

plt.plot([[x] for x in np.linspace(1, 3, 100)], Y, color='r')
plt.show()

loss_ad = []
for epoch in range(1000):
    out = model_ad(X)
    ad.zero_grad()
    loss = loss_function(out, y)
    loss.backward()
    ad.step()

    print('epoch={}, loss={}'.format(epoch, loss.data.numpy()))
    loss_ad.append(loss.data.numpy())

plt.plot(loss_sgd, 'b')
plt.plot(loss_ad, 'r')
plt.show()