模型训练相关

思路：

1. 导入数据集（对数据集转换为张量）
2. 加载数据集（使数据集成为可以进行迭代）
3. 搭建卷积模型
4. 进行模型训练（每训练一轮查看一次在测试集上的准确率）
5. 使用tensorboard进行可视化
6. 保存训练后的模型
7. 加载训练好的模型进行测试.

选择的模型结构

训练模型和评估模型

在conda命令行输入
tensorboard --logdir=“tensorboard --logdir=D:\student\ai-study\02框架学习\logs\mnist”
打开可视化面板


在测试集上的准确率不断上升

训练和评估完整代码

"""
@author:Lunau
@file:022_mnist.py
@time:2024/08/07
@任务：使用pytorch对mnist数据集进行训练和测试
"""
import torch
import torchvision
import time
from torch import nn
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter

"""
@root:存放数据集的目录
@train:为True表示是作为训练集
@transforms:导入数据集的同时进行预处理
@download:为True表示从网络下载模型
"""
# 导入数据集 这里没有归一化
train_data = torchvision.datasets.MNIST("./dataset/MNIST", train=True, transform=
                                        torchvision.transforms.ToTensor(), download=True)
test_data = torchvision.datasets.MNIST("./dataset/MNIST", train=False, transform=
                                       torchvision.transforms.ToTensor(), download=True)


# # 加载数据集,方便进行迭代
train_dataloader = DataLoader(dataset=train_data, batch_size=64)
test_dataloader = DataLoader(dataset=test_data, batch_size=64)
# img, target = test_data[0]
# print(img)
# print(target)

# 构建卷积层
class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.model = nn.Sequential(
            nn.Conv2d(in_channels=1, out_channels=10, kernel_size=5, padding=0, stride=1),
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(in_channels=10, out_channels=20, kernel_size=5, padding=0, stride=1),
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=2),
            nn.Flatten(),
            nn.Linear(in_features=320, out_features=10)
        )

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

# 创建卷积模型
model = Model()

# 损失函数 交叉熵
loss_fn = nn.CrossEntropyLoss()

# 优化器
learning_rate = 0.01
optimizer = torch.optim.SGD(params=model.parameters(), lr=learning_rate)


# 可视化
writer = SummaryWriter("./logs/mnist")

# 训练网络的参数
total_train_step = 0  # 训练次数
total_test_step = 0  # 测试次数

# 训练
def train():
    # 训练步骤开始
    model.train()
    global total_train_step
    for data in train_dataloader:
        imgs, targets = data
        outputs = model(imgs)
        loss = loss_fn(outputs, targets)  # 计算当前损失
        # 优化器进行优化
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        total_train_step += 1
        if total_train_step % 100 ==0:
            writer.add_scalar("train_loss", loss, total_train_step) # 可视化每轮的损失
            print(f"训练次数：{total_train_step}， Loss：{loss}")

def test():
    global total_test_step

    model.eval()
    total_test_loss = 0
    total_accuracy = 0
    test_data_len = len(test_data)

    with torch.no_grad():
        for data in test_dataloader:
            imgs, targets = data
            output = model(imgs)
            loss = loss_fn(output, targets)
            total_test_loss +=loss
            accuracy = (output.argmax(1) == targets).sum().item() # 计算出正确的次数
            total_accuracy+=accuracy
        total_accuracy = total_accuracy / test_data_len
    # 在整体测试集上的损失
    print(f"整体测试损失Loss:{total_test_loss}")
    # 整体测试的正确率
    print(f"整体测试的正确率acc：{total_accuracy}")

    writer.add_scalar("test_loss", total_test_loss, total_test_step)
    writer.add_scalar("test_acc", total_accuracy, total_test_step)
    total_test_step +=1
    return total_accuracy



if __name__ == '__main__':

    epoch = 10  # 训练的轮数
    for i in range(epoch):
        print(f"第{i + 1}轮训练开始")
        train()
        test()
    torch.save(model, "mnist1.pth")

测试模型

测试的照片

测试代码

"""
@author:Lunau
@file:023_mnist_test.py
@time:2024/08/07
"""
import cv2
import torch
import torchvision
import torchvision.transforms as transforms
from PIL import Image
from torch import nn

# 测试模型

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.model = nn.Sequential(
            nn.Conv2d(in_channels=1, out_channels=10, kernel_size=5, padding=0, stride=1),
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(in_channels=10, out_channels=20, kernel_size=5, padding=0, stride=1),
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=2),
            nn.Flatten(),
            nn.Linear(in_features=320, out_features=10)
        )

    def forward(self, x):
        x = self.model(x)
        return x
# 测试单张照片
def test_one_image():
    image_path = "./images/mnist_3.jpg"
    image = Image.open(image_path)
    print(image)
    image = image.convert('L')
    image.show()

    # 图片进行转换
    transform = torchvision.transforms.Compose([
        torchvision.transforms.Resize((28, 28)),
        torchvision.transforms.ToTensor()
    ])

    image = transform(image)
    print(image.shape)

    # 加载模型 若模型是在gpu训练出来，需要在cpu上运行需要进行一个映射
    model = torch.load("./mnist1.pth")
    image = torch.reshape(image, (1, 1, 28, 28))  # 转换一下尺寸，为输入要求的尺寸
    # 测试
    model.eval()
    with torch.no_grad():
        output = model(image)
    print(output)
    print(f"手写数字是：{output.argmax(1).item()}")

test_one_image()