Pytorch实现卷积神经网络CNN

本文使用Pytorch实现卷积神经网络CNN，仅为Pytorch新入坑者提供帮助，避免入门到放弃；
数据集为MNIST如文中代码所示；
最后，恭喜各位入坑，在调参的路上越走越远！

import torch
from torch.autograd import Variable
import torch.utils.data as Data
import torchvision  # 数据库模块
import torch.nn as nn
import matplotlib.pyplot as plt


LR = 0.001  # 学习率
BATCH_SIZE = 50  # 表示每次选取50个样本作训练
EPOCH = 1  # epoch表示整个数据集重复训练次数
DOWNLOAD_MNIST = False  # 是否下载MNIST

train_data = torchvision.datasets.MNIST(root='./mnist',
					train=True,  # 表示这是训练集
                                       	transform=torchvision.transforms.ToTensor(),  # 原始数据是array数组，转换为tensor,同时进行归一化
                                        download=DOWNLOAD_MNIST)

# 打印图片

print(train_data.data.size())  # 打印训练集的大小
print(train_data.targets.size())  # 打印训练集标签的大小
‘’使用imshow()函数加载训练集中的图片’‘’
plt.imshow(train_data.data[0].numpy(), cmap='gray')  
plt.title('%i' % train_data.targets[0])
plt.show()

# 训练
train_loader = Data.DataLoader(dataset=train_data, batch_size=BATCH_SIZE,
                              			 shuffle=True, num_workers=2)  # 批量化（包装）处理数据集

test_data = torchvision.datasets.MNIST(root='./mnist', train=False)

‘’‘#测试集shape from (2000, 28, 28) to (2000, 1, 28, 28), value in range(0,1)’‘’
test_x = torch.unsqueeze(test_data.data, dim=[:2000].type(torch.FloatTensor).cuda()/255.  

‘’‘#这里使用cuda()表示使用调用GPU加速跑测试集/训练集’‘’
test_y = test_data.targets[:2000].cuda()  

# 搭建CNN网络
if __name__ == '__main__':
    class CNN(torch.nn.Module):  # 定义一个神经网络层
        def __init__(self):  # 定义初始化
            super(CNN, self).__init__()  # 继承初始层
            self.conv1 = torch.nn.Sequential(
                nn.Conv2d(in_channels=1,
                          out_channels=16,  # 卷积核个数
                          kernel_size=5,  # 卷积核大小
                          stride=1,  # 卷积核移动步长
                          padding=2,  # 填充数目
                          ),  #-->卷积后的图片形状大小(28,28,16)
                nn.ReLU(),
                nn.MaxPool2d(kernel_size=2),  #-->池化后的图片形状大小(14,14,16)
            )
            self.conv2 = nn.Sequential(
                nn.Conv2d(16, 32, 5, 1, 2),  #-->第二次卷积后的图片形状大小(14,14,32)
                nn.ReLU(),
                nn.MaxPool2d(kernel_size=2)  #-->第二次池化后的图片形状大小(7,7,32)
                )
            self.out = nn.Linear(32*7*7, 10)  # 全连接层

        def forward(self, x):
            x = self.conv1(x)
            x = self.conv2(x)
            x = x.view(x.size(0), -1)  # 数据展平为多维后的图片形状大小(batch size,7,7,32)
            output = self.out(x)
            return output

    cnn = CNN()
    print(cnn)  # 打印CNN网络
    cnn.cuda()  # CNN模型使用GPU加速

    optimizer = torch.optim.Adam(cnn.parameters(), lr=LR， betas=(0.9, 0.99))  # Adam优化器
    loss_func = nn.CrossEntropyLoss()  # 损失函数

    # 训练模型
    for epoch in range(EPOCH):
        for step, (b_x, b_y) in enumerate(train_loader):
            output = cnn(b_x.cuda())  # CNN预测输出
            loss = loss_func(output, b_y.cuda())
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()

            if step % 50 == 0:
                test_output = cnn(test_x.cuda())
                test_pred = torch.max(test_output, 1)[1].cuda().data
                test_accuracy = float((test_pred == test_y).sum().item()) / float(test_y.size(0))
                print('Epoch:', epoch, 'Train loss:', loss.data.cpu().numpy(), 'Test Accuracy:', test_accuracy)

    # 观察前20个数据误差
    test_output = cnn(test_x[:20])
    pred_y = torch.max(test_output, 1)[1].cuda().data
    print(pred_y.cpu().numpy(), 'prediction number')
    print(test_y[:20].cpu().numpy(), 'real number')

前20张测试集图片标签预测结果如下所示：

Epoch: 0 Train loss: 2.3024046 Test Accuracy: 0.1915
Epoch: 0 Train loss: 0.7746874 Test Accuracy: 0.832
Epoch: 0 Train loss: 0.23315571 Test Accuracy: 0.884
Epoch: 0 Train loss: 0.2868251 Test Accuracy: 0.9175
Epoch: 0 Train loss: 0.3143016 Test Accuracy: 0.9255
Epoch: 0 Train loss: 0.11868002 Test Accuracy: 0.938
Epoch: 0 Train loss: 0.10169793 Test Accuracy: 0.941
Epoch: 0 Train loss: 0.2048641 Test Accuracy: 0.9525
Epoch: 0 Train loss: 0.12582631 Test Accuracy: 0.9675
Epoch: 0 Train loss: 0.03774266 Test Accuracy: 0.9625
Epoch: 0 Train loss: 0.051590152 Test Accuracy: 0.967
Epoch: 0 Train loss: 0.007484264 Test Accuracy: 0.9675
Epoch: 0 Train loss: 0.15514685 Test Accuracy: 0.9705
Epoch: 0 Train loss: 0.03676407 Test Accuracy: 0.97
Epoch: 0 Train loss: 0.035673436 Test Accuracy: 0.967
Epoch: 0 Train loss: 0.06010488 Test Accuracy: 0.9715
Epoch: 0 Train loss: 0.13796066 Test Accuracy: 0.9765
Epoch: 0 Train loss: 0.09808495 Test Accuracy: 0.974
Epoch: 0 Train loss: 0.016342957 Test Accuracy: 0.977
Epoch: 0 Train loss: 0.11043023 Test Accuracy: 0.9745
Epoch: 0 Train loss: 0.038324982 Test Accuracy: 0.9795
Epoch: 0 Train loss: 0.11415327 Test Accuracy: 0.9805
Epoch: 0 Train loss: 0.15161397 Test Accuracy: 0.981
Epoch: 0 Train loss: 0.11598789 Test Accuracy: 0.97
[7 2 1 0 4 1 4 9 5 9 0 6 9 0 1 5 9 7 3 4] prediction number
[7 2 1 0 4 1 4 9 5 9 0 6 9 0 1 5 9 7 3 4] real number