Pytorch入门案例---线性回归

import torch
from torch import nn
import numpy as np
import torch.utils.data as Data
from torch.nn import init

# 使得模型的可复现性
torch.manual_seed(1)
# 设置默认的数据格式
torch.set_default_tensor_type('torch.FloatTensor')

# 1.数据处理
num_inputs = 2
num_examples = 1000

true_w = [2, -3.4]
true_b = 4.2

features = torch.tensor(np.random.normal(0, 1, (num_examples, num_inputs)), dtype=torch.float)
labels = true_w[0] * features[:, 0] + true_w[1] * features[:, 1] + true_b

# 增加数据的噪声, 模拟真实数据
labels += torch.tensor(np.random.normal(0, 0.01, size=labels.size()), dtype=torch.float)

# 2.读取数据
batch_size = 10

dataset = Data.TensorDataset(features, labels)
# 查看数据的存储类型 [*dataset]
data_iter = Data.DataLoader(dataset, batch_size, shuffle=True, num_workers=2)

# 3.定义网络
class LinearNet(nn.Module):
    def __init__(self, n_feature):
        super(LinearNet, self).__init__()
        self.linear = nn.Linear(n_feature, 1)
    
    def forward(self, x):
        return self.linear(x)
net = LinearNet(feature)

# 初始化网络的参数(有多种初始化的方法,但是一般偏值设置为零)
for name, param in net.named_parameters():
    if name == "linear.weight":
        #权重
        init.normal_(param, mean=0.0, std=0.01)
    else:
        #偏值
        init.constant_(param, val=0.0)

# 4.训练
# 超参数
lr = 0.01
batch_size = 10
num_epochs = 5
# 定义损失
critrion = nn.MSELoss()
# 定义优化函数
optimizer = t.optim.SGD(net.parameters(), lr=lr)

for epoch in range(num_epochs):
    for X, y in data_iter:
        out = net(X)
        loss = critrion(out, y.view(-1, 1))
        # 梯度清零
        optimizer.zero_grad()
        loss.backward()
        # 梯度更新
        optimizer.step()
    print("epcoch:%d, loss:%f"%(epoch, loss.item()))

• 结果

epcoch:0, loss:0.861575
epcoch:1, loss:0.015179
epcoch:2, loss:0.000336
epcoch:3, loss:0.000168
epcoch:4, loss:0.000151