Pytorch学习笔记（模型训练）

模型训练

在同一个包下创建train.py和model.py，按照步骤先从数据处理，模型架构搭建，训练测试，统计损失，如下面代码所示

1. train.py

import torch.optim
import torchvision
from torch import nn
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter

from model import NNN

# 1. 准备数据集
train_data = torchvision.datasets.CIFAR10("./data", train=True, transform=torchvision.transforms.ToTensor(),
                                          download=True)
test_data = torchvision.datasets.CIFAR10("./data", train=False, transform=torchvision.transforms.ToTensor(),
                                         download=True)

train_data_size = len(train_data)
test_data_size = len(test_data)
print(f"训练数据集的长度：{train_data_size}")
print(f"测试数据集的长度：{test_data_size}")

# 2. 利用DataLoader 加载数据集
train_dataloader = DataLoader(train_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)

# 3. 搭建神经网络
# 引入model.py
nnn = NNN()

# 4. 创建损失函数loss
loss_fn = nn.CrossEntropyLoss()  # 交叉熵

# 5. 优化器
learning_rate = 0.01
optimizer = torch.optim.SGD(nnn.parameters(), lr=learning_rate)  # 随机梯度下降

# 6. 设置训练网络的一些参数
total_train_step = 0  # 记录训练次数
total_test_step = 0  # 训练测试次数
epoch = 10  # 训练轮数

# 补充tensorboard
writer = SummaryWriter("../logs")

# 开始训练
for i in range(epoch):
    print(f"--------第{i+1}轮训练开始--------")
    # 训练
    nnn.train()
    for data in train_dataloader:
        imgs, targets = data
        outputs = nnn(imgs)
        loss = loss_fn(outputs, targets)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        total_train_step += 1
        if total_train_step % 100 == 0:
            print(f"训练次数：{total_train_step}---loss：{loss.item()}")
            writer.add_scalar("train_loss", loss.item(), total_train_step)

    # 测试
    nnn.eval()
    total_test_loss = 0  # 总体的误差
    total_accuracy = 0  # 总体的正确率
    with torch.no_grad():
        for data in test_dataloader:
            imgs, targets = data
            outputs = nnn(imgs)
            loss = loss_fn(outputs, targets)
            total_test_loss += loss.item()
            accuracy = (outputs.argmax(1) == targets).sum()
            total_accuracy += accuracy
    print(f"整体测试集上的loss：{total_test_loss}")
    print(f"整体测试集上的准确率：{total_accuracy/test_data_size}")
    writer.add_scalar("test_loss", total_test_loss, total_test_step)
    writer.add_scalar("total_accuracy", total_accuracy/test_data_size, total_test_step)
    total_test_step += 1

    # 保存每一轮训练的模型
    torch.save(nnn, f"nnn_{i+1}.pth")
    print("模式已保存")


writer.close()
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
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85

2. model.py

import torch
from torch import nn


# 搭建神经网络
class NNN(nn.Module):
    def __init__(self):
        super(NNN, self).__init__()
        self.model = nn.Sequential(
            nn.Conv2d(3, 32, 5, stride=1, padding=2),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(32, 32, 5, stride=1, padding=2),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(32, 64, 5, stride=1, padding=2),
            nn.MaxPool2d(2),
            nn.Flatten(),
            nn.Linear(1024, 64),
            nn.Linear(64, 10)
        )

    def forward(self, x):
        x = self.model(x)
        return x


if __name__ == '__main__':
    nnn = NNN()
    input = torch.ones((64, 3, 32, 32))
    output = nnn(input)
    print(output.shape)

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

运行train.py后可以通过启动tensorboard进行查看我们的loss情况，损失是不断下降的。

补充argmax函数的使用
我们模型预测处理的是概率，我们需要使用argmax函数还得到预测的结果，就是选出概率最大的，上面测试准确率的计算使用到了。
简单代码示例：

import torch
# 模型输出的概率
outputs = torch.tensor([[0.1, 0.3],
                        [0.7, 0.2]])
# 真实的分类
targets = torch.tensor([[1, 1]])
# 对概率进行预测
preds = outputs.argmax(1)  # 1:横向比较 0：竖向比较

# 预测与真实进行比较
print(preds == targets)
print((preds == targets).sum().item())  # 统计正确的个数

1
2
3
4
5
6
7
8
9
10
11
12
13

输出：

tensor([[ True, False]])
1
1
2