神经网络：pytorch线性回归的简洁实现

线性回归的简洁实现

例子：用线性回归模型拟合带有服从-0.5 到 0.5 的均匀分布噪声的正弦函数

import numpy as np
from matplotlib import pyplot as plt
import torch
from torch.utils import data
from torch import nn

生成数据集: 正弦函数+随机噪声

num_observations = 100
x = np.linspace(-3, 3, num_observations)
y = np.sin(x) + np.random.uniform(-0.5, 0.5, num_observations)

构建数据迭代器

def load_array(data_arrays, batch_size, is_train=True):
    dataset = data.TensorDataset(*data_arrays)
    return data.DataLoader(dataset, batch_size, shuffle=is_train)

简单处理原始数据, 初始化数据迭代器

batch_size = 10
features = torch.from_numpy(x).type(torch.float32).reshape(-1, 1)
labels = torch.from_numpy(y).type(torch.float32).reshape(-1, 1)

data_iter = load_array((features, labels), batch_size)

next(iter(data_iter))

output:

[tensor([[-2.5152],
         [ 2.3939],
         [ 1.2424],
         [-0.5152],
         [ 1.6667],
         [ 2.5758],
         [-0.3939],
         [ 1.8485],
         [ 0.6364],
         [ 0.0909]]),
 tensor([[-0.6717],
         [ 0.3632],
         [ 1.0844],
         [-0.1079],
         [ 1.2082],
         [ 0.1642],
         [-0.1016],
         [ 0.6608],
         [ 0.5274],
         [-0.3599]])]

定义模型
用框架中提前准备好的构件很方便

net = nn.Sequential(nn.Linear(1,1))

初始化模型参数

net[0].weight.data.normal_(0, 0.01)
net[0].bias.data.fill_(0)

tensor([0.])

定义损失函数

loss = nn.MSELoss()

定义优化方法

trainer = torch.optim.SGD(net.parameters(), lr = 0.005)

开始训练

num_epochs = 100
for epoch in range(num_epochs):
    for X, y in data_iter:
        l = loss(net(X), y)
        trainer.zero_grad()
        l.backward()
        trainer.step()
    l = loss(net(features), labels)
    print(f'epoch {epoch + 1}, loss {l:f}, w {net[0].weight.data[0, 0]}, b {net[0].bias.data[0]}')

output

epoch 1, loss 0.410211, w 0.07609143853187561, b 0.0010609857272356749
epoch 2, loss 0.328852, w 0.140553817152977, b 0.001841965364292264
...
epoch 99, loss 0.236213, w 0.31469690799713135, b 0.009414778091013432
epoch 100, loss 0.236213, w 0.3147128224372864, b 0.009223075583577156

展示训练结果

_y = net(features)
plt.title("linear model")
plt.xlabel("x")
plt.ylabel("y")
plt.plot(features.numpy(), labels.numpy(), 'ob')
plt.plot(features.numpy(), _y.detach().numpy(), 'r')
plt.show()

完整代码：pytorch线性回归的简洁实现
参考内容：动手学深度学习v2 # 李沐