手把手教你实现PyTorch的MNIST数据集-巨人网络通讯

手把手教你实现PyTorch的MNIST数据集

概述

MNIST 包含 0~9 的手写数字, 共有 60000 个训练集和 10000 个测试集. 数据的格式为单通道 28*28 的灰度图.

获取数据

def get_data():
    """获取数据"""

    # 获取测试集
    train = torchvision.datasets.MNIST(root="./data", train=True, download=True,
                                       transform=torchvision.transforms.Compose([
                                           torchvision.transforms.ToTensor(),  # 转换成张量
                                           torchvision.transforms.Normalize((0.1307,), (0.3081,))  # 标准化
                                       ]))
    train_loader = DataLoader(train, batch_size=batch_size)  # 分割测试集

    # 获取测试集
    test = torchvision.datasets.MNIST(root="./data", train=False, download=True,
                                      transform=torchvision.transforms.Compose([
                                          torchvision.transforms.ToTensor(),  # 转换成张量
                                          torchvision.transforms.Normalize((0.1307,), (0.3081,))  # 标准化
                                      ]))
    test_loader = DataLoader(test, batch_size=batch_size)  # 分割训练

    # 返回分割好的训练集和测试集
    return train_loader, test_loader

网络模型

class Model(torch.nn.Module):
    def __init__(self):
        super(Model, self).__init__()

        # 卷积层
        self.conv1 = torch.nn.Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1))
        self.conv2 = torch.nn.Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))

        # Dropout层
        self.dropout1 = torch.nn.Dropout(0.25)
        self.dropout2 = torch.nn.Dropout(0.5)

        # 全连接层
        self.fc1 = torch.nn.Linear(9216, 128)
        self.fc2 = torch.nn.Linear(128, 10)

    def forward(self, x):
        """前向传播"""
        
        # [b, 1, 28, 28] => [b, 32, 26, 26]
        out = self.conv1(x)
        out = F.relu(out)
        
        # [b, 32, 26, 26] => [b, 64, 24, 24]
        out = self.conv2(out)
        out = F.relu(out)

        # [b, 64, 24, 24] => [b, 64, 12, 12]
        out = F.max_pool2d(out, 2)
        out = self.dropout1(out)
        
        # [b, 64, 12, 12] => [b, 64 * 12 * 12] => [b, 9216]
        out = torch.flatten(out, 1)
        
        # [b, 9216] => [b, 128]
        out = self.fc1(out)
        out = F.relu(out)

        # [b, 128] => [b, 10]
        out = self.dropout2(out)
        out = self.fc2(out)

        output = F.log_softmax(out, dim=1)

        return output

train 函数

def train(model, epoch, train_loader):
    """训练"""

    # 训练模式
    model.train()

    # 迭代
    for step, (x, y) in enumerate(train_loader):
        # 加速
        if use_cuda:
            model = model.cuda()
            x, y = x.cuda(), y.cuda()

        # 梯度清零
        optimizer.zero_grad()

        output = model(x)

        # 计算损失
        loss = F.nll_loss(output, y)

        # 反向传播
        loss.backward()

        # 更新梯度
        optimizer.step()

        # 打印损失
        if step % 50 == 0:
            print('Epoch: {}, Step {}, Loss: {}'.format(epoch, step, loss))

test 函数

def test(model, test_loader):
    """测试"""
    
    # 测试模式
    model.eval()

    # 存放正确个数
    correct = 0

    with torch.no_grad():
        for x, y in test_loader:

            # 加速
            if use_cuda:
                model = model.cuda()
                x, y = x.cuda(), y.cuda()

            # 获取结果
            output = model(x)

            # 预测结果
            pred = output.argmax(dim=1, keepdim=True)

            # 计算准确个数
            correct += pred.eq(y.view_as(pred)).sum().item()

    # 计算准确率
    accuracy = correct / len(test_loader.dataset) * 100

    # 输出准确
    print("Test Accuracy: {}%".format(accuracy))

main 函数

def main():
    # 获取数据
    train_loader, test_loader = get_data()
    
    # 迭代
    for epoch in range(iteration_num):
        print("\n================ epoch: {} ================".format(epoch))
        train(network, epoch, train_loader)
        test(network, test_loader)

完整代码:

import torch
import torchvision
import torch.nn.functional as F
from torch.utils.data import DataLoader
class Model(torch.nn.Module):
    def __init__(self):
        super(Model, self).__init__()

        # 卷积层
        self.conv1 = torch.nn.Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1))
        self.conv2 = torch.nn.Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))

        # Dropout层
        self.dropout1 = torch.nn.Dropout(0.25)
        self.dropout2 = torch.nn.Dropout(0.5)

        # 全连接层
        self.fc1 = torch.nn.Linear(9216, 128)
        self.fc2 = torch.nn.Linear(128, 10)

    def forward(self, x):
        """前向传播"""
        
        # [b, 1, 28, 28] => [b, 32, 26, 26]
        out = self.conv1(x)
        out = F.relu(out)
        
        # [b, 32, 26, 26] => [b, 64, 24, 24]
        out = self.conv2(out)
        out = F.relu(out)

        # [b, 64, 24, 24] => [b, 64, 12, 12]
        out = F.max_pool2d(out, 2)
        out = self.dropout1(out)
        
        # [b, 64, 12, 12] => [b, 64 * 12 * 12] => [b, 9216]
        out = torch.flatten(out, 1)
        
        # [b, 9216] => [b, 128]
        out = self.fc1(out)
        out = F.relu(out)

        # [b, 128] => [b, 10]
        out = self.dropout2(out)
        out = self.fc2(out)

        output = F.log_softmax(out, dim=1)

        return output


# 定义超参数
batch_size = 64  # 一次训练的样本数目
learning_rate = 0.0001  # 学习率
iteration_num = 5  # 迭代次数
network = Model()  # 实例化网络
print(network)  # 调试输出网络结构
optimizer = torch.optim.Adam(network.parameters(), lr=learning_rate)  # 优化器

# GPU 加速
use_cuda = torch.cuda.is_available()
print("是否使用 GPU 加速:", use_cuda)


def get_data():
    """获取数据"""

    # 获取测试集
    train = torchvision.datasets.MNIST(root="./data", train=True, download=True,
                                       transform=torchvision.transforms.Compose([
                                           torchvision.transforms.ToTensor(),  # 转换成张量
                                           torchvision.transforms.Normalize((0.1307,), (0.3081,))  # 标准化
                                       ]))
    train_loader = DataLoader(train, batch_size=batch_size)  # 分割测试集

    # 获取测试集
    test = torchvision.datasets.MNIST(root="./data", train=False, download=True,
                                      transform=torchvision.transforms.Compose([
                                          torchvision.transforms.ToTensor(),  # 转换成张量
                                          torchvision.transforms.Normalize((0.1307,), (0.3081,))  # 标准化
                                      ]))
    test_loader = DataLoader(test, batch_size=batch_size)  # 分割训练

    # 返回分割好的训练集和测试集
    return train_loader, test_loader


def train(model, epoch, train_loader):
    """训练"""

    # 训练模式
    model.train()

    # 迭代
    for step, (x, y) in enumerate(train_loader):
        # 加速
        if use_cuda:
            model = model.cuda()
            x, y = x.cuda(), y.cuda()

        # 梯度清零
        optimizer.zero_grad()

        output = model(x)

        # 计算损失
        loss = F.nll_loss(output, y)

        # 反向传播
        loss.backward()

        # 更新梯度
        optimizer.step()

        # 打印损失
        if step % 50 == 0:
            print('Epoch: {}, Step {}, Loss: {}'.format(epoch, step, loss))


def test(model, test_loader):
    """测试"""

    # 测试模式
    model.eval()

    # 存放正确个数
    correct = 0

    with torch.no_grad():
        for x, y in test_loader:

            # 加速
            if use_cuda:
                model = model.cuda()
                x, y = x.cuda(), y.cuda()

            # 获取结果
            output = model(x)

            # 预测结果
            pred = output.argmax(dim=1, keepdim=True)

            # 计算准确个数
            correct += pred.eq(y.view_as(pred)).sum().item()

    # 计算准确率
    accuracy = correct / len(test_loader.dataset) * 100

    # 输出准确
    print("Test Accuracy: {}%".format(accuracy))


def main():
    # 获取数据
    train_loader, test_loader = get_data()

    # 迭代
    for epoch in range(iteration_num):
        print("\n================ epoch: {} ================".format(epoch))
        train(network, epoch, train_loader)
        test(network, test_loader)

if __name__ == "__main__":
    main()

输出结果:

Model(
(conv1): Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1))
(conv2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))
(dropout1): Dropout(p=0.25, inplace=False)
(dropout2): Dropout(p=0.5, inplace=False)
(fc1): Linear(in_features=9216, out_features=128, bias=True)
(fc2): Linear(in_features=128, out_features=10, bias=True)
)
是否使用 GPU 加速: True

================ epoch: 0 ================
Epoch: 0, Step 0, Loss: 2.3131277561187744
Epoch: 0, Step 50, Loss: 1.0419045686721802
Epoch: 0, Step 100, Loss: 0.6259541511535645
Epoch: 0, Step 150, Loss: 0.7194482684135437
Epoch: 0, Step 200, Loss: 0.4020516574382782
Epoch: 0, Step 250, Loss: 0.6890509128570557
Epoch: 0, Step 300, Loss: 0.28660136461257935
Epoch: 0, Step 350, Loss: 0.3277580738067627
Epoch: 0, Step 400, Loss: 0.2750288248062134
Epoch: 0, Step 450, Loss: 0.28428223729133606
Epoch: 0, Step 500, Loss: 0.3514065444469452
Epoch: 0, Step 550, Loss: 0.23386947810649872
Epoch: 0, Step 600, Loss: 0.25338059663772583
Epoch: 0, Step 650, Loss: 0.1743898093700409
Epoch: 0, Step 700, Loss: 0.35752204060554504
Epoch: 0, Step 750, Loss: 0.17575909197330475
Epoch: 0, Step 800, Loss: 0.20604261755943298
Epoch: 0, Step 850, Loss: 0.17389622330665588
Epoch: 0, Step 900, Loss: 0.3188241124153137
Test Accuracy: 96.56%

================ epoch: 1 ================
Epoch: 1, Step 0, Loss: 0.23558208346366882
Epoch: 1, Step 50, Loss: 0.13511177897453308
Epoch: 1, Step 100, Loss: 0.18823786079883575
Epoch: 1, Step 150, Loss: 0.2644936144351959
Epoch: 1, Step 200, Loss: 0.145077645778656
Epoch: 1, Step 250, Loss: 0.30574971437454224
Epoch: 1, Step 300, Loss: 0.2386859953403473
Epoch: 1, Step 350, Loss: 0.08346735686063766
Epoch: 1, Step 400, Loss: 0.10480977594852448
Epoch: 1, Step 450, Loss: 0.07280707359313965
Epoch: 1, Step 500, Loss: 0.20928426086902618
Epoch: 1, Step 550, Loss: 0.20455852150917053
Epoch: 1, Step 600, Loss: 0.10085935145616531
Epoch: 1, Step 650, Loss: 0.13476189970970154
Epoch: 1, Step 700, Loss: 0.19087043404579163
Epoch: 1, Step 750, Loss: 0.0981522724032402
Epoch: 1, Step 800, Loss: 0.1961515098810196
Epoch: 1, Step 850, Loss: 0.041140712797641754
Epoch: 1, Step 900, Loss: 0.250461220741272
Test Accuracy: 98.03%

================ epoch: 2 ================
Epoch: 2, Step 0, Loss: 0.09572553634643555
Epoch: 2, Step 50, Loss: 0.10370486229658127
Epoch: 2, Step 100, Loss: 0.17737184464931488
Epoch: 2, Step 150, Loss: 0.1570713371038437
Epoch: 2, Step 200, Loss: 0.07462178170681
Epoch: 2, Step 250, Loss: 0.18744900822639465
Epoch: 2, Step 300, Loss: 0.09910508990287781
Epoch: 2, Step 350, Loss: 0.08929706364870071
Epoch: 2, Step 400, Loss: 0.07703761011362076
Epoch: 2, Step 450, Loss: 0.10133732110261917
Epoch: 2, Step 500, Loss: 0.1314031481742859
Epoch: 2, Step 550, Loss: 0.10394387692213058
Epoch: 2, Step 600, Loss: 0.11612939089536667
Epoch: 2, Step 650, Loss: 0.17494803667068481
Epoch: 2, Step 700, Loss: 0.11065669357776642
Epoch: 2, Step 750, Loss: 0.061209067702293396
Epoch: 2, Step 800, Loss: 0.14715790748596191
Epoch: 2, Step 850, Loss: 0.03930797800421715
Epoch: 2, Step 900, Loss: 0.18030673265457153
Test Accuracy: 98.46000000000001%

================ epoch: 3 ================
Epoch: 3, Step 0, Loss: 0.09266342222690582
Epoch: 3, Step 50, Loss: 0.0414913073182106
Epoch: 3, Step 100, Loss: 0.2152961939573288
Epoch: 3, Step 150, Loss: 0.12287424504756927
Epoch: 3, Step 200, Loss: 0.13468700647354126
Epoch: 3, Step 250, Loss: 0.11967387050390244
Epoch: 3, Step 300, Loss: 0.11301510035991669
Epoch: 3, Step 350, Loss: 0.037447575479745865
Epoch: 3, Step 400, Loss: 0.04699449613690376
Epoch: 3, Step 450, Loss: 0.05472381412982941
Epoch: 3, Step 500, Loss: 0.09839300811290741
Epoch: 3, Step 550, Loss: 0.07964356243610382
Epoch: 3, Step 600, Loss: 0.08182843774557114
Epoch: 3, Step 650, Loss: 0.05514759197831154
Epoch: 3, Step 700, Loss: 0.13785190880298615
Epoch: 3, Step 750, Loss: 0.062480345368385315
Epoch: 3, Step 800, Loss: 0.120387002825737
Epoch: 3, Step 850, Loss: 0.04458726942539215
Epoch: 3, Step 900, Loss: 0.17119190096855164
Test Accuracy: 98.55000000000001%

================ epoch: 4 ================
Epoch: 4, Step 0, Loss: 0.08094145357608795
Epoch: 4, Step 50, Loss: 0.05615215748548508
Epoch: 4, Step 100, Loss: 0.07766406238079071
Epoch: 4, Step 150, Loss: 0.07915271818637848
Epoch: 4, Step 200, Loss: 0.1301635503768921
Epoch: 4, Step 250, Loss: 0.12118984013795853
Epoch: 4, Step 300, Loss: 0.073218435049057
Epoch: 4, Step 350, Loss: 0.04517696052789688
Epoch: 4, Step 400, Loss: 0.08493026345968246
Epoch: 4, Step 450, Loss: 0.03904269263148308
Epoch: 4, Step 500, Loss: 0.09386837482452393
Epoch: 4, Step 550, Loss: 0.12583576142787933
Epoch: 4, Step 600, Loss: 0.09053893387317657
Epoch: 4, Step 650, Loss: 0.06912104040384293
Epoch: 4, Step 700, Loss: 0.1502612829208374
Epoch: 4, Step 750, Loss: 0.07162325084209442
Epoch: 4, Step 800, Loss: 0.10512275993824005
Epoch: 4, Step 850, Loss: 0.028180215507745743
Epoch: 4, Step 900, Loss: 0.08492615073919296
Test Accuracy: 98.69%