ML_LLM_AI►Pytorch

004-neural network 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
# PyTorch 神经网络构建指南

## 神经网络构建基础

PyTorch 使用 `torch.nn` 程序包构建神经网络：
- 基于 `autograd` 定义和区分模型
- `nn.Module` 包含：
  - 多个网络层
  - `forward(input)` 方法（返回输出）

## 案例1：定义网络

```python
import torch
import torch.nn as nn
import torch.nn.functional as F

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        # 定义网络层
        self.conv1 = nn.Conv2d(1, 6, 3)     # 1输入通道，6输出通道，3x3卷积核
        self.conv2 = nn.Conv2d(6, 16, 3)    # 6输入通道，16输出通道，3x3卷积核
        self.fc1 = nn.Linear(16*6*6, 120)   # 全连接层 (16*6*6 -> 120)
        self.fc2 = nn.Linear(120, 84)       # 全连接层 (120 -> 84)
        self.fc3 = nn.Linear(84, 10)        # 全连接层 (84 -> 10)
    
    def forward(self, x):
        # 前向传播
        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))  # 卷积+ReLU+2x2最大池化
        x = F.max_pool2d(F.relu(self.conv2(x)), 2)       # 卷积+ReLU+2x2最大池化
        x = x.view(-1, self.num_flat_features(x))         # 展平
        x = F.relu(self.fc1(x))                           # 全连接+ReLU
        x = F.relu(self.fc2(x))                           # 全连接+ReLU
        x = self.fc3(x)                                   # 输出层
        return x
    
    def num_flat_features(self, x):
        # 计算展平后的特征数量
        size = x.size()[1:]  # 除批量维度外的所有维度
        num_features = 1
        for s in size:
            num_features *= s
        return num_features

# 实例化网络
net = Net()
print(net)

重要说明：

  1. 自动反向传播：只需定义 forward 函数，backward 会自动通过 autograd 计算梯度
  2. 小批量支持：torch.nn 仅支持小批量输入，不支持单个样本
    • 例如：nn.Conv2d 需要 4D 张量输入格式：nSamples × nChannels × Height × Width
  3. 网络层说明：
    • nn.Conv2d：二维卷积层
    • nn.Linear：全连接层（仿射变换 y = Wx + b）
    • F.relu：ReLU激活函数
    • F.max_pool2d：二维最大池化
ML_LLM_AI►Pytorch

005-model training code 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
# PyTorch 图像分类器训练指南

## CIFAR100 图像分类训练

### 1. 加载并规范化数据集

```python
import torch
import torchvision
import torchvision.transforms as transforms

# 定义数据转换
transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])

# 加载训练集
trainset = torchvision.datasets.CIFAR100(
    root='./data', 
    train=True,
    download=True, 
    transform=transform
)
trainloader = torch.utils.data.DataLoader(
    trainset, 
    batch_size=4,
    shuffle=True, 
    num_workers=2
)

# 加载测试集
testset = torchvision.datasets.CIFAR100(
    root='./data', 
    train=False,
    download=True, 
    transform=transform
)
testloader = torch.utils.data.DataLoader(
    testset, 
    batch_size=4,
    shuffle=False, 
    num_workers=2
)

2. 定义卷积神经网络

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
import torch.nn as nn
import torch.nn.functional as F

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(3, 6, 5)          # 3输入通道，6输出通道，5x5卷积核
        self.pool = nn.MaxPool2d(2, 2)           # 2x2最大池化
        self.conv2 = nn.Conv2d(6, 16, 5)         # 6输入通道，16输出通道，5x5卷积核
        self.fc1 = nn.Linear(16 * 5 * 5, 120)    # 全连接层
        self.fc2 = nn.Linear(120, 84)            # 全连接层
        self.fc3 = nn.Linear(84, 100)            # 输出层(CIFAR100有100类)

    def forward(self, x):
        x = self.pool(F.relu(self.conv1(x)))
        x = self.pool(F.relu(self.conv2(x)))
        x = x.view(-1, 16 * 5 * 5)               # 展平
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x

net = Net()

3. 定义损失函数和优化器

1
2
3
4
5
6
7
8
import torch.optim as optim

criterion = nn.CrossEntropyLoss()  # 交叉熵损失
optimizer = optim.SGD(             # SGD优化器
    net.parameters(), 
    lr=0.001, 
    momentum=0.9
)

4. 训练网络

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
for epoch in range(2):  # 遍历数据集多次
    running_loss = 0.0
    for i, data in enumerate(trainloader, 0):
        inputs, labels = data
        
        # 梯度清零
        optimizer.zero_grad()
        
        # 前向传播+反向传播+优化
        outputs = net(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()
        
        # 打印统计信息
        running_loss += loss.item()
        if i % 2000 == 1999:  # 每2000个小批量打印一次
            print('[%d, %5d] loss: %.3f' % 
                 (epoch + 1, i + 1, running_loss / 2000))
            running_loss = 0.0

print('Finished Training')

# 保存模型
PATH = './cifar_net.pth'
torch.save(net.state_dict(), PATH)

5. 测试网络性能

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
# 测试整个数据集
correct = 0
total = 0
with torch.no_grad():
    for data in testloader:
        images, labels = data
        outputs = net(images)
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

print('Accuracy on 10000 test images: %d %%' % (100 * correct / total))

# 按类别测试准确率
class_correct = list(0. for _ in range(100))
class_total = list(0. for _ in range(100))
with torch.no_grad():
    for data in testloader:
        images, labels = data
        outputs = net(images)
        _, predicted = torch.max(outputs, 1)
        c = (predicted == labels).squeeze()
        for i in range(4):
            label = labels[i]
            class_correct[label] += c[i].item()
            class_total[label] += 1

for i in range(100):
    print('Accuracy of %5s : %2d %%' % (
        classes[i], 100 * class_correct[i] / class_total[i]))

线性回归示例

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
import torch
from torch import nn

# 1. 准备数据
x = torch.tensor([[1.0], [2.0], [3.0]])
y = torch.tensor([[3.0], [6.0], [9.0]])

# 2. 定义模型
class LinearModel(nn.Module):
    def __init__(self):
        super(LinearModel, self).__init__()
        self.linear = nn.Linear(1, 1)  # 输入输出维度都是1
        
    def forward(self, x):
        return self.linear(x)

model = LinearModel()

# 3. 定义损失和优化器
criterion = nn.MSELoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

# 4. 训练循环
for epoch in range(1000):
    y_pred = model(x)          # 前向传播
    loss = criterion(y_pred, y) # 计算损失
    
    optimizer.zero_grad()      # 梯度清零
    loss.backward()            # 反向传播
    optimizer.step()           # 更新参数
    
    if epoch % 100 == 0:
        print(f'Epoch {epoch}, Loss: {loss.item()}')

# 打印训练结果
print('w =', model.linear.weight.item())
print('b =', model.linear.bias.item())