第二次作业：卷积神经网络 part01

视频学习

机器学习的数学基础

特征向量形象化的描述：对一个矩阵施加线性变换后，使矩阵发生尺度变化而不改变方向。

秩形象化的描述：秩序，复杂度，一个数据分布很容易被捕捉，则秩小，很难被捕捉，则秩大。

数据降维：只保留前R个较大奇异值及其对应的特征向量（较大奇异值包含了矩阵的主要信息）。

低秩近似：保留决定数据分布的最主要的模式/方向（丢弃的可能是噪声或其他不关键的信息）。

概率/函数形式的统一：

问题补充：

逐层训练时，在训练下一层时，会冻结上一层的参数。

逐层预训练初始化参数是为了更好的初始化，使其落到比较好的区域里面。

策略设计：训练误差->泛化误差

免费午餐定理：

奥卡姆剃刀原理：

“如无必要，勿增实体”， 即“简单有效原理”。如果多种模型能够同等程度地符合一个问题的观测结果，应选择其中使用假设最少的->最简单的模型。

欠拟合和过拟合的解决办法：

频率学派VS贝叶斯学派：

频率学派VS机器学习方法：

卷积神经网络基本组成结构

卷积神经网络的应用：分类、检索、检测、分割人脸识别、人脸验证、人脸表情识别、图像生成图像风格转换、自动驾驶。

传统神经网络VS卷积神经网络：

​ 深度学习的三部曲：

​ 1.搭建神经网络结构

​ 2.找到一个合适的损失函数

​ 3.找到一个合适的优化函数，更新参数

​ 损失函数：

​ 全连接网络处理图像的问题：参数太多：权重矩阵的参数太多->过拟合

​ 卷积络的解决方式：局部关联，参数共享

​ 两者的相同之处：都有卷积层、激活层、池化层和全连接层

卷积：

​

池化：

全连接：

卷积神经网络的典型架构

Alexnet：

ZFNet：

VGG：

GoogleNet：

GoogleNet：Naive Inception 的计算复杂度过高

GoogleNet：Inception 插入1*1卷积核进行降维

GoogleNet：Inception V3 对V2的参数数量进行降低 增加非线性激活函数，使其表征能力更强，训练更快。

ResNet：

面试小问题：ResNet50层以下和50层以上有啥区别：50层以上由bottleNeck，50层以下没有

MNIST数据集分类

深度神经网络的特性：

• 很多层: compositionality
• 卷积: locality + stationarity of images
• 池化: Invariance of object class to translations

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
import matplotlib.pyplot as plt
import numpy

# 计算模型当中含有多少参数
def get_n_params(model):
  np = 0
  for p in list(model.parameters()):
    np += p.nelement()
  return np

# 使用GPU来进行训练
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

1.加载数据（MINST）

Pytorch里面含有MINST、CIFAR10等常用的数据集，调用 torchvision.datasets 即可把这些数据由远程下载到本地

MNIST的使用方法：

torchvision.datasets.MNIST(root, train=True, transform=None, target_transform=None, download=False)

• root 为数据集下载到本地后的根目录，包括 training.pt 和 test.pt 文件
• train，如果设置为True，从training.pt创建数据集，否则从test.pt创建。
• download，如果设置为True, 从互联网下载数据并放到root文件夹下
• transform, 一种函数或变换，输入PIL图片，返回变换之后的数据。
• target_transform 一种函数或变换，输入目标，进行变换。

另外值得注意的是，DataLoader是一个比较重要的类，提供的常用操作有：batch_size(每个batch的大小), shuffle(是否进行随机打乱顺序的操作), num_workers(加载数据的时候使用几个子进程)

input_size = 28*28 # MINST上的图像是28*28大的
output_size = 10  # 类别是0-9的数字 因此分为十类

train_loader = torch.utils.data.DataLoader(
    datasets.MNIST('./data', train = True, download = True, transform = transforms.Compose(
        [transforms.ToTensor(),
         transforms.Normalize((0.1307,),(0.3081,))])),
    batch_size = 64, shuffle = True)

train_loader = torch.utils.data.DataLoader(
    datasets.MNIST('./data', train = False, transform = transforms.Compose(
        [transforms.ToTensor(),
         transforms.Normalize((0.1307,),(0.3081,))])),
    batch_size = 1000, shuffle = True)

显示数据集中的部分图像

plt.figure(figsize = (8, 5))
for i in range(20):
    plt.subplot(4, 5, i + 1)
    image, _ = train_loader.dataset.__getitem__(i)
    plt.imshow(image.squeeze().numpy(),'gray')
    plt.axis('off');

2.创建网络

定义网络时，需要继承nn.Module，并实现它的forward方法，把网络中具有可学习参数的层放在构造函数init中。

只要在nn.Module的子类中定义了forward函数，backward函数就会自动被实现(利用autograd)。

class FC2Layer(nn.Module):
    def __init__(self, input_size, n_hidden, output_size):
        # nn.Module子类的函数必须在构造函数中执行父类的构造函数
        # 下式等价于nn.Module.__init__(self)        
        super(FC2Layer, self).__init__()
        self.input_size = input_size
        # 这里直接用 Sequential 就定义了网络，注意要和下面 CNN 的代码区分开
        self.network = nn.Sequential(
            nn.Linear(input_size, n_hidden), 
            nn.ReLU(), 
            nn.Linear(n_hidden, n_hidden), 
            nn.ReLU(), 
            nn.Linear(n_hidden, output_size), 
            nn.LogSoftmax(dim=1)
        )
    def forward(self, x):
        # view一般出现在model类的forward函数中，用于改变输入或输出的形状
        # x.view(-1, self.input_size) 的意思是多维的数据展成二维
        # 代码指定二维数据的列数为 input_size=784，行数 -1 表示我们不想算，电脑会自己计算对应的数字
        # 在 DataLoader 部分，我们可以看到 batch_size 是64，所以得到 x 的行数是64
        # 大家可以加一行代码：print(x.cpu().numpy().shape)
        # 训练过程中，就会看到 (64, 784) 的输出，和我们的预期是一致的

        # forward 函数的作用是，指定网络的运行过程，这个全连接网络可能看不啥意义，
        # 下面的CNN网络可以看出 forward 的作用。
        x = x.view(-1, self.input_size)
        return self.network(x)
    


class CNN(nn.Module):
    def __init__(self, input_size, n_feature, output_size):
        # 执行父类的构造函数，所有的网络都要这么写
        super(CNN, self).__init__()
        # 下面是网络里典型结构的一些定义，一般就是卷积和全连接
        # 池化、ReLU一类的不用在这里定义
        self.n_feature = n_feature
        self.conv1 = nn.Conv2d(in_channels=1, out_channels=n_feature, kernel_size=5)
        self.conv2 = nn.Conv2d(n_feature, n_feature, kernel_size=5)
        self.fc1 = nn.Linear(n_feature*4*4, 50)
        self.fc2 = nn.Linear(50, 10)    
    
    # 下面的 forward 函数，定义了网络的结构，按照一定顺序，把上面构建的一些结构组织起来
    # 意思就是，conv1, conv2 等等的，可以多次重用
    def forward(self, x, verbose=False):
        x = self.conv1(x)
        x = F.relu(x)
        x = F.max_pool2d(x, kernel_size=2)
        x = self.conv2(x)
        x = F.relu(x)
        x = F.max_pool2d(x, kernel_size=2)
        x = x.view(-1, self.n_feature*4*4)
        x = self.fc1(x)
        x = F.relu(x)
        x = self.fc2(x)
        x = F.log_softmax(x, dim=1)
        return x

定义训练和测试函数

# 训练函数
def train(model):
    model.train()
    # 主里从train_loader里，64个样本一个batch为单位提取样本进行训练
    for batch_idx, (data, target) in enumerate(train_loader):
        # 把数据送到GPU中
        data, target = data.to(device), target.to(device)

        optimizer.zero_grad()
        output = model(data)
        loss = F.nll_loss(output, target)
        loss.backward()
        optimizer.step()
        if batch_idx % 100 == 0:
            print('Train: [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                batch_idx * len(data), len(train_loader.dataset),
                100. * batch_idx / len(train_loader), loss.item()))


def test(model):
    model.eval()
    test_loss = 0
    correct = 0
    for data, target in test_loader:
        # 把数据送到GPU中
        data, target = data.to(device), target.to(device)
        # 把数据送入模型，得到预测结果
        output = model(data)
        # 计算本次batch的损失，并加到 test_loss 中
        test_loss += F.nll_loss(output, target, reduction='sum').item()
        # get the index of the max log-probability，最后一层输出10个数，
        # 值最大的那个即对应着分类结果，然后把分类结果保存在 pred 里
        pred = output.data.max(1, keepdim=True)[1]
        # 将 pred 与 target 相比，得到正确预测结果的数量，并加到 correct 中
        # 这里需要注意一下 view_as ，意思是把 target 变成维度和 pred 一样的意思                                                
        correct += pred.eq(target.data.view_as(pred)).cpu().sum().item()

    test_loss /= len(test_loader.dataset)
    accuracy = 100. * correct / len(test_loader.dataset)
    print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
        test_loss, correct, len(test_loader.dataset),
        accuracy))

3.在小型全连接网络上训练（Fully-connected network）

n_hidden = 8 # number of hidden units

model_fnn = FC2Layer(input_size, n_hidden, output_size)
model_fnn.to(device)
optimizer = optim.SGD(model_fnn.parameters(), lr=0.01, momentum=0.5)
print('Number of parameters: {}'.format(get_n_params(model_fnn)))

train(model_fnn)
test(model_fnn)

Number of parameters: 6442
Train: [0/60000 (0%)]	Loss: 2.287285
Train: [6400/60000 (11%)]	Loss: 1.967152
Train: [12800/60000 (21%)]	Loss: 1.261107
Train: [19200/60000 (32%)]	Loss: 0.959287
Train: [25600/60000 (43%)]	Loss: 0.747636
Train: [32000/60000 (53%)]	Loss: 0.476292
Train: [38400/60000 (64%)]	Loss: 0.652557
Train: [44800/60000 (75%)]	Loss: 0.392811
Train: [51200/60000 (85%)]	Loss: 0.443054
Train: [57600/60000 (96%)]	Loss: 0.350913

Test set: Average loss: 0.4373, Accuracy: 8715/10000 (87%)

4、在卷积神经网络上训练

定义的CNN和全连接网络拥有相同数量的模型参数

# Training settings 
n_features = 6 # number of feature maps

model_cnn = CNN(input_size, n_features, output_size)
model_cnn.to(device)
optimizer = optim.SGD(model_cnn.parameters(), lr=0.01, momentum=0.5)
print('Number of parameters: {}'.format(get_n_params(model_cnn)))

train(model_cnn)
test(model_cnn)

Number of parameters: 6422
Train: [0/60000 (0%)]	Loss: 2.260017
Train: [6400/60000 (11%)]	Loss: 1.045880
Train: [12800/60000 (21%)]	Loss: 0.482751
Train: [19200/60000 (32%)]	Loss: 0.486722
Train: [25600/60000 (43%)]	Loss: 0.354835
Train: [32000/60000 (53%)]	Loss: 0.192605
Train: [38400/60000 (64%)]	Loss: 0.172775
Train: [44800/60000 (75%)]	Loss: 0.121670
Train: [51200/60000 (85%)]	Loss: 0.128894
Train: [57600/60000 (96%)]	Loss: 0.177849

Test set: Average loss: 0.1605, Accuracy: 9497/10000 (95%)

很明显相同参数下的CNN网络优于简单的全连接网络，原因是CNN网络可以更好的挖掘图像中的信息，主要通过两种手段：卷积和池化。

5.打乱像素顺序再次在两个网络上训练和测试

# 这里解释一下 torch.randperm 函数，给定参数n，返回一个从0到n-1的随机整数排列
perm = torch.randperm(784)
plt.figure(figsize=(8, 4))
for i in range(10):
    image, _ = train_loader.dataset.__getitem__(i)
    # permute pixels
    image_perm = image.view(-1, 28*28).clone()
    image_perm = image_perm[:, perm]
    image_perm = image_perm.view(-1, 1, 28, 28)
    plt.subplot(4, 5, i + 1)
    plt.imshow(image.squeeze().numpy(), 'gray')
    plt.axis('off')
    plt.subplot(4, 5, i + 11)
    plt.imshow(image_perm.squeeze().numpy(), 'gray')
    plt.axis('off')

重新定义训练与测试函数，我们写了两个函数 train_perm 和 test_perm，分别对应着加入像素打乱顺序的训练函数与测试函数。

与之前的训练与测试函数基本上完全相同，只是对 data 加入了打乱顺序操作。

# 对每个 batch 里的数据，打乱像素顺序的函数
def perm_pixel(data, perm):
    # 转化为二维矩阵
    data_new = data.view(-1, 28*28)
    # 打乱像素顺序
    data_new = data_new[:, perm]
    # 恢复为原来4维的 tensor
    data_new = data_new.view(-1, 1, 28, 28)
    return data_new

# 训练函数
def train_perm(model, perm):
    model.train()
    for batch_idx, (data, target) in enumerate(train_loader):
        data, target = data.to(device), target.to(device)
        # 像素打乱顺序
        data = perm_pixel(data, perm)

        optimizer.zero_grad()
        output = model(data)
        loss = F.nll_loss(output, target)
        loss.backward()
        optimizer.step()
        if batch_idx % 100 == 0:
            print('Train: [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                batch_idx * len(data), len(train_loader.dataset),
                100. * batch_idx / len(train_loader), loss.item()))

# 测试函数
def test_perm(model, perm):
    model.eval()
    test_loss = 0
    correct = 0
    for data, target in test_loader:
        data, target = data.to(device), target.to(device)

        # 像素打乱顺序
        data = perm_pixel(data, perm)

        output = model(data)
        test_loss += F.nll_loss(output, target, reduction='sum').item()
        pred = output.data.max(1, keepdim=True)[1]                                            
        correct += pred.eq(target.data.view_as(pred)).cpu().sum().item()

    test_loss /= len(test_loader.dataset)
    accuracy = 100. * correct / len(test_loader.dataset)
    print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
        test_loss, correct, len(test_loader.dataset),
        accuracy))

小型全连接网络：

perm = torch.randperm(784)
n_hidden = 8 # number of hidden units

model_fnn = FC2Layer(input_size, n_hidden, output_size)
model_fnn.to(device)
optimizer = optim.SGD(model_fnn.parameters(), lr=0.01, momentum=0.5)
print('Number of parameters: {}'.format(get_n_params(model_fnn)))

train_perm(model_fnn, perm)
test_perm(model_fnn, perm)

Number of parameters: 6442
Train: [0/60000 (0%)]	Loss: 2.294211
Train: [6400/60000 (11%)]	Loss: 1.820978
Train: [12800/60000 (21%)]	Loss: 1.060106
Train: [19200/60000 (32%)]	Loss: 0.781592
Train: [25600/60000 (43%)]	Loss: 0.718575
Train: [32000/60000 (53%)]	Loss: 0.595289
Train: [38400/60000 (64%)]	Loss: 0.603694
Train: [44800/60000 (75%)]	Loss: 0.590031
Train: [51200/60000 (85%)]	Loss: 0.745443
Train: [57600/60000 (96%)]	Loss: 0.731628

Test set: Average loss: 0.5092, Accuracy: 8448/10000 (84%)

CNN：

perm = torch.randperm(784)
n_features = 6 # number of feature maps

model_cnn = CNN(input_size, n_features, output_size)
model_cnn.to(device)
optimizer = optim.SGD(model_cnn.parameters(), lr=0.01, momentum=0.5)
print('Number of parameters: {}'.format(get_n_params(model_cnn)))

train_perm(model_cnn, perm)
test_perm(model_cnn, perm)

Number of parameters: 6422
Train: [0/60000 (0%)]	Loss: 2.302913
Train: [6400/60000 (11%)]	Loss: 2.259660
Train: [12800/60000 (21%)]	Loss: 2.158767
Train: [19200/60000 (32%)]	Loss: 1.658526
Train: [25600/60000 (43%)]	Loss: 1.302927
Train: [32000/60000 (53%)]	Loss: 1.099095
Train: [38400/60000 (64%)]	Loss: 0.969859
Train: [44800/60000 (75%)]	Loss: 0.835230
Train: [51200/60000 (85%)]	Loss: 0.598763
Train: [57600/60000 (96%)]	Loss: 0.552054

Test set: Average loss: 0.6426, Accuracy: 7941/10000 (79%)

小结：打乱像素顺序之后，小型全连接网络的性能上没有发生明显的变化，而CNN性能很明显的下降，这是因为CNN会利用像素的局部关系，但是打乱顺序之后，这些像素间的关系将无法被利用。

CIFAR10数据集分类

CIFAR10数据集包含十个类别：‘airplane’, ‘automobile’, ‘bird’, ‘cat’, ‘deer’, ‘dog’, ‘frog’, ‘horse’, ‘ship’, ‘truck’，其中的图像大小为33232，也就是RGB的3层颜色通道，每层通道中的尺寸为32*32。

PyTorch 创建了一个叫做 totchvision 的包，该包含有支持加载类似Imagenet，CIFAR10，MNIST 等公共数据集的数据加载模块 torchvision.datasets 和支持加载图像数据数据转换模块 torch.utils.data.DataLoader。

首先，加载并归一化 CIFAR10 使用 torchvision 。torchvision 数据集的输出是范围在[0,1]之间的 PILImage，我们将他们转换成归一化范围为[-1,1]之间的张量 Tensors。

import torch
import torchvision
import torchvision.transforms as transforms
import matplotlib.pyplot as plt
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim

# 使用GPU训练
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
# input[channel] = (input[channel] - mean[channel]) / std[channel]

# 注意下面代码中：训练的 shuffle 是 True，测试的 shuffle 是 false
# 训练时可以打乱顺序增加多样性，测试是没有必要
trainset = torchvision.datasets.CIFAR10(root='./data', train=True,
                                        download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=64,
                                          shuffle=True, num_workers=2)

testset = torchvision.datasets.CIFAR10(root='./data', train=False,
                                       download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=8,
                                         shuffle=False, num_workers=2)

classes = ('plane', 'car', 'bird', 'cat',
           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')

Downloading https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz to ./data/cifar-10-python.tar.gz
97%
165552128/170498071 [00:02<00:00, 81789133.73it/s]
Extracting ./data/cifar-10-python.tar.gz to ./data
Files already downloaded and verified

CIFAR10中的一些图片：

def imshow(img):
    plt.figure(figsize=(8,8))
    img = img / 2 + 0.5     # 转换到 [0,1] 之间
    npimg = img.numpy()
    plt.imshow(np.transpose(npimg, (1, 2, 0)))
    plt.show()

# 得到一组图像
images, labels = iter(trainloader).next()
# 展示图像
imshow(torchvision.utils.make_grid(images))
# 展示第一行图像的标签
for j in range(8):
    print(classes[labels[j]])

定义网络、损失函数和优化器：

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(3, 6, 5)
        self.pool = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(6, 16, 5)
        self.fc1 = nn.Linear(16 * 5 * 5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)

    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)
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x

# 网络放到GPU上
net = Net().to(device)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(net.parameters(), lr=0.001)

训练网络：

for epoch in range(10):  # 重复多轮训练
    for i, (inputs, labels) in enumerate(trainloader):
        inputs = inputs.to(device)
        labels = labels.to(device)
        # 优化器梯度归零
        optimizer.zero_grad()
        # 正向传播 +　反向传播 + 优化 
        outputs = net(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()
        # 输出统计信息
        if i % 100 == 0:   
            print('Epoch: %d Minibatch: %5d loss: %.3f' %(epoch + 1, i + 1, loss.item()))

print('Finished Training')

Epoch: 1 Minibatch:     1 loss: 2.319
Epoch: 1 Minibatch:   101 loss: 1.739
Epoch: 1 Minibatch:   201 loss: 1.641
Epoch: 1 Minibatch:   301 loss: 1.571
Epoch: 1 Minibatch:   401 loss: 1.370
Epoch: 1 Minibatch:   501 loss: 1.565
Epoch: 1 Minibatch:   601 loss: 1.452
Epoch: 1 Minibatch:   701 loss: 1.448
Epoch: 2 Minibatch:     1 loss: 1.543
Epoch: 2 Minibatch:   101 loss: 1.447
Epoch: 2 Minibatch:   201 loss: 1.441
Epoch: 2 Minibatch:   301 loss: 1.320
Epoch: 2 Minibatch:   401 loss: 1.407
Epoch: 2 Minibatch:   501 loss: 1.245
Epoch: 2 Minibatch:   601 loss: 1.348
Epoch: 2 Minibatch:   701 loss: 1.199
Epoch: 3 Minibatch:     1 loss: 1.171
Epoch: 3 Minibatch:   101 loss: 1.180
Epoch: 3 Minibatch:   201 loss: 1.178
Epoch: 3 Minibatch:   301 loss: 1.251
Epoch: 3 Minibatch:   401 loss: 1.328
Epoch: 3 Minibatch:   501 loss: 1.357
Epoch: 3 Minibatch:   601 loss: 1.302
Epoch: 3 Minibatch:   701 loss: 1.203
Epoch: 4 Minibatch:     1 loss: 1.332
Epoch: 4 Minibatch:   101 loss: 1.207
Epoch: 4 Minibatch:   201 loss: 1.074
Epoch: 4 Minibatch:   301 loss: 1.312
Epoch: 4 Minibatch:   401 loss: 1.105
Epoch: 4 Minibatch:   501 loss: 0.974
Epoch: 4 Minibatch:   601 loss: 0.966
Epoch: 4 Minibatch:   701 loss: 1.284
Epoch: 5 Minibatch:     1 loss: 1.244
Epoch: 5 Minibatch:   101 loss: 1.175
Epoch: 5 Minibatch:   201 loss: 1.098
Epoch: 5 Minibatch:   301 loss: 1.008
Epoch: 5 Minibatch:   401 loss: 1.174
Epoch: 5 Minibatch:   501 loss: 1.090
Epoch: 5 Minibatch:   601 loss: 1.228
Epoch: 5 Minibatch:   701 loss: 1.377
Epoch: 6 Minibatch:     1 loss: 1.021
Epoch: 6 Minibatch:   101 loss: 0.756
Epoch: 6 Minibatch:   201 loss: 1.045
Epoch: 6 Minibatch:   301 loss: 1.175
Epoch: 6 Minibatch:   401 loss: 1.147
Epoch: 6 Minibatch:   501 loss: 1.155
Epoch: 6 Minibatch:   601 loss: 1.173
Epoch: 6 Minibatch:   701 loss: 1.143
Epoch: 7 Minibatch:     1 loss: 1.027
Epoch: 7 Minibatch:   101 loss: 1.187
Epoch: 7 Minibatch:   201 loss: 1.093
Epoch: 7 Minibatch:   301 loss: 0.758
Epoch: 7 Minibatch:   401 loss: 0.922
Epoch: 7 Minibatch:   501 loss: 1.218
Epoch: 7 Minibatch:   601 loss: 1.177
Epoch: 7 Minibatch:   701 loss: 0.895
Epoch: 8 Minibatch:     1 loss: 0.918
Epoch: 8 Minibatch:   101 loss: 0.902
Epoch: 8 Minibatch:   201 loss: 0.876
Epoch: 8 Minibatch:   301 loss: 0.927
Epoch: 8 Minibatch:   401 loss: 0.854
Epoch: 8 Minibatch:   501 loss: 1.201
Epoch: 8 Minibatch:   601 loss: 0.909
Epoch: 8 Minibatch:   701 loss: 1.001
Epoch: 9 Minibatch:     1 loss: 0.982
Epoch: 9 Minibatch:   101 loss: 1.132
Epoch: 9 Minibatch:   201 loss: 0.758
Epoch: 9 Minibatch:   301 loss: 0.896
Epoch: 9 Minibatch:   401 loss: 1.013
Epoch: 9 Minibatch:   501 loss: 1.014
Epoch: 9 Minibatch:   601 loss: 1.013
Epoch: 9 Minibatch:   701 loss: 0.965
Epoch: 10 Minibatch:     1 loss: 0.825
Epoch: 10 Minibatch:   101 loss: 0.971
Epoch: 10 Minibatch:   201 loss: 0.841
Epoch: 10 Minibatch:   301 loss: 0.870
Epoch: 10 Minibatch:   401 loss: 1.079
Epoch: 10 Minibatch:   501 loss: 0.831
Epoch: 10 Minibatch:   601 loss: 0.741
Epoch: 10 Minibatch:   701 loss: 1.007
Finished Training

从测试集里面取出8张图片：

# 得到一组图像
images, labels = iter(testloader).next()
# 展示图像
imshow(torchvision.utils.make_grid(images))
# 展示图像的标签
for j in range(8):
    print(classes[labels[j]])

把图片输入模型，看CNN把图片识别成什么：

outputs = net(images.to(device))
_, predicted = torch.max(outputs, 1)

# 展示预测的结果
for j in range(8):
    print(classes[predicted[j]])

cat
car
ship
ship
deer
frog
dog
frog

从结果中可以看出有几个识别错了

查看CNN在整个数据集上的表现：

correct = 0
total = 0

for data in testloader:
    images, labels = data
    images, labels = images.to(device), labels.to(device)
    outputs = net(images)
    _, predicted = torch.max(outputs.data, 1)
    total += labels.size(0)
    correct += (predicted == labels).sum().item()

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

Accuracy of the network on the 10000 test images: 63 %

准确率大概为63%

使用 VGG16 对 CIFAR10 分类

1.定义dataloader

CIFAR10是3x32x32的，所以transform定义是三通道的

import torch
import torchvision
import torchvision.transforms as transforms
import matplotlib.pyplot as plt
import numpy as np
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim

# 使用GPU训练，可以在菜单 "代码执行工具" -> "更改运行时类型" 里进行设置
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

transform_train = transforms.Compose([
    transforms.RandomCrop(32, padding=4),
    transforms.RandomHorizontalFlip(),
    transforms.ToTensor(),
    transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))])

transform_test = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))])

trainset = torchvision.datasets.CIFAR10(root='./data', train=True,  download=True, transform=transform_train)
testset  = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform_test)

trainloader = torch.utils.data.DataLoader(trainset, batch_size=128, shuffle=True, num_workers=2)
testloader = torch.utils.data.DataLoader(testset, batch_size=128, shuffle=False, num_workers=2)

classes = ('plane', 'car', 'bird', 'cat',
           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')

Files already downloaded and verified
Files already downloaded and verified

2.定义VGG网络

class VGG(nn.Module):
    def __init__(self):
        super(VGG, self).__init__()
        self.cfg = [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M']
        self.features = self._make_layers(self.cfg)
        self.classifier = nn.Linear(2048, 10)

    def forward(self, x):
        out = self.features(x)
        out = out.view(out.size(0), -1)
        out = self.classifier(out)
        return out

    def _make_layers(self, cfg):
        layers = []
        in_channels = 3
        for x in cfg:
            if x == 'M':
                layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
            else:
                layers += [nn.Conv2d(in_channels, x, kernel_size=3, padding=1),
                           nn.BatchNorm2d(x),
                           nn.ReLU(inplace=True)]
                in_channels = x
        layers += [nn.AvgPool2d(kernel_size=1, stride=1)]
        return nn.Sequential(*layers)

初始化网络，根据实际需要，修改分类层。因为 tiny-imagenet 是对200类图像分类，这里把输出修改为200。

# 网络放到GPU上
net = VGG().to(device)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(net.parameters(), lr=0.001)

3.网络训练

for epoch in range(10):  # 重复多轮训练
    for i, (inputs, labels) in enumerate(trainloader):
        inputs = inputs.to(device)
        labels = labels.to(device)
        # 优化器梯度归零
        optimizer.zero_grad()
        # 正向传播 +　反向传播 + 优化 
        outputs = net(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()
        # 输出统计信息
        if i % 100 == 0:   
            print('Epoch: %d Minibatch: %5d loss: %.3f' %(epoch + 1, i + 1, loss.item()))

print('Finished Training')

Epoch: 1 Minibatch:     1 loss: 2.472
Epoch: 1 Minibatch:   101 loss: 1.387
Epoch: 1 Minibatch:   201 loss: 1.332
Epoch: 1 Minibatch:   301 loss: 1.125
Epoch: 2 Minibatch:     1 loss: 1.009
Epoch: 2 Minibatch:   101 loss: 1.108
Epoch: 2 Minibatch:   201 loss: 1.140
Epoch: 2 Minibatch:   301 loss: 1.039
Epoch: 3 Minibatch:     1 loss: 0.986
Epoch: 3 Minibatch:   101 loss: 0.879
Epoch: 3 Minibatch:   201 loss: 0.646
Epoch: 3 Minibatch:   301 loss: 0.586
Epoch: 4 Minibatch:     1 loss: 0.676
Epoch: 4 Minibatch:   101 loss: 0.757
Epoch: 4 Minibatch:   201 loss: 0.639
Epoch: 4 Minibatch:   301 loss: 0.713
Epoch: 5 Minibatch:     1 loss: 0.795
Epoch: 5 Minibatch:   101 loss: 0.746
Epoch: 5 Minibatch:   201 loss: 0.585
Epoch: 5 Minibatch:   301 loss: 0.686
Epoch: 6 Minibatch:     1 loss: 0.494
Epoch: 6 Minibatch:   101 loss: 0.556
Epoch: 6 Minibatch:   201 loss: 0.506
Epoch: 6 Minibatch:   301 loss: 0.596
Epoch: 7 Minibatch:     1 loss: 0.587
Epoch: 7 Minibatch:   101 loss: 0.421
Epoch: 7 Minibatch:   201 loss: 0.470
Epoch: 7 Minibatch:   301 loss: 0.473
Epoch: 8 Minibatch:     1 loss: 0.677
Epoch: 8 Minibatch:   101 loss: 0.560
Epoch: 8 Minibatch:   201 loss: 0.505
Epoch: 8 Minibatch:   301 loss: 0.493
Epoch: 9 Minibatch:     1 loss: 0.362
Epoch: 9 Minibatch:   101 loss: 0.425
Epoch: 9 Minibatch:   201 loss: 0.301
Epoch: 9 Minibatch:   301 loss: 0.340
Epoch: 10 Minibatch:     1 loss: 0.428
Epoch: 10 Minibatch:   101 loss: 0.368
Epoch: 10 Minibatch:   201 loss: 0.335
Epoch: 10 Minibatch:   301 loss: 0.396
Finished Training

4.测试正确率

correct = 0
total = 0

for data in testloader:
    images, labels = data
    images, labels = images.to(device), labels.to(device)
    outputs = net(images)
    _, predicted = torch.max(outputs.data, 1)
    total += labels.size(0)
    correct += (predicted == labels).sum().item()

print('Accuracy of the network on the 10000 test images: %.2f %%' % (
    100 * correct / total))

Accuracy of the network on the 10000 test images: 84.48 %

使用VGG模型进行猫狗大战

import numpy as np
import matplotlib.pyplot as plt
import os
import torch
import torch.nn as nn
import torchvision
from torchvision import models,transforms,datasets
import time
import json


# 判断是否存在GPU设备
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print('Using gpu: %s ' % torch.cuda.is_available())

Using gpu: True 

下载数据

! wget https://static.leiphone.com/cat_dog.rar

--2020-07-31 13:02:05--  https://static.leiphone.com/cat_dog.rar
Resolving static.leiphone.com (static.leiphone.com)... 47.246.24.229, 47.246.24.227, 47.246.24.228, ...
Connecting to static.leiphone.com (static.leiphone.com)|47.246.24.229|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 546904884 (522M) [application/x-rar-compressed]
Saving to: ‘cat_dog.rar’

cat_dog.rar         100%[===================>] 521.57M  19.4MB/s    in 28s     

2020-07-31 13:02:33 (18.9 MB/s) - ‘cat_dog.rar’ saved [546904884/546904884]

因为AI研习社提供的数据集是rar，所以需要安装rarfile库

pip install rarfile

Collecting rarfile
  Downloading https://files.pythonhosted.org/packages/59/66/d2a475dce12051fa93d80c07cb1aea663e6ab15afc2c2973ab53cd14a0f0/rarfile-3.3.tar.gz (135kB)
     |████████████████████████████████| 143kB 3.5MB/s 
Building wheels for collected packages: rarfile
  Building wheel for rarfile (setup.py) ... done
  Created wheel for rarfile: filename=rarfile-3.3-py2.py3-none-any.whl size=24969 sha256=9076dc220e263686553095f5e2106dd924e55817a70bbf7dc9d4d3a7349b89d2
  Stored in directory: /root/.cache/pip/wheels/77/9b/af/37bc95a3007ad325d678785dc65f6ee48bba34ecf0019cf9be
Successfully built rarfile
Installing collected packages: rarfile
Successfully installed rarfile-3.3

解压cat_dog文件

import rarfile
path = "cat_dog.rar"
path2 = "/content/"
rf = rarfile.RarFile(path) 
rf.extractall(path2)

数据处理

datasets 是 torchvision 中的一个包，可以用做加载图像数据。它可以以多线程（multi-thread）的形式从硬盘中读取数据，使用 mini-batch 的形式，在网络训练中向 GPU 输送。在使用CNN处理图像时，需要进行预处理。图片将被整理成 的大小，同时还将进行归一化处理。torchvision 支持对输入数据进行一些复杂的预处理/变换 （normalization, cropping, flipping, jittering 等）。

normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])

vgg_format = transforms.Compose([
                transforms.CenterCrop(224),
                transforms.ToTensor(),
                normalize,
            ])

data_dir = './cat_dog/'

dsets = {x: datasets.ImageFolder(os.path.join(data_dir, x), vgg_format)
         for x in ['train', 'valid']}

dset_sizes = {x: len(dsets[x]) for x in ['train', 'valid']}
dset_classes = dsets['train'].classes

直接运行代码发现报错

Found 0 files in subfolders of: ./cat_dog/train
Supported extensions are: .jpg,.jpeg,.png,.ppm,.bmp,.pgm,.tif,.tiff,.webp

搜索之后线下载该数据集的文件存储结构跟Pytorch的规范格式不一致，所以要进行预处理

mkdir cat_dog/val/Dog

mkdir cat_dog/val/Cat

mkdir cat_dog/train/Cat

mkdir cat_dog/train/Dog

mkdir cat_dog/test/test

mv cat_dog/val/dog* cat_dog/val/Dog/

mv cat_dog/val/cat* cat_dog/val/Cat/

mv cat_dog/train/cat* cat_dog/train/Cat/

mv cat_dog/train/dog* cat_dog/train/Dog/

mv cat_dog/test/*.jpg cat_dog/test/test

normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])

vgg_format_train = transforms.Compose([
                transforms.RandomRotation(30),# 随机旋转
                transforms.CenterCrop(224),
                transforms.ToTensor(),
                normalize,
                
            ])

vgg_format = transforms.Compose([
                transforms.CenterCrop(224),
                transforms.ToTensor(),
                normalize,
            ])

data_dir = './cat_dog/'

# 利用ImageFolder进行分类文件夹加载
# 两种加载数据集的方法
dsets = {x: datasets.ImageFolder(os.path.join(data_dir, x), vgg_format)
         for x in ['train', 'val']}

tsets = {y: datasets.ImageFolder(os.path.join(data_dir, y), vgg_format)
        for y in ['test']}
dset_classes = dsets['train'].classes
dset_sizes = {x: len(dsets[x]) for x in ['train', 'val']}

# 通过下面代码可以查看 dsets 的一些属性

print(dsets['train'].classes)
print(dsets['train'].class_to_idx)
print(dsets['train'].imgs[:5])
print('dset_sizes: ', dset_sizes)

['Cat', 'Dog']
{'Cat': 0, 'Dog': 1}
[('./cat_dog/train/Cat/cat_0.jpg', 0), ('./cat_dog/train/Cat/cat_1.jpg', 0), ('./cat_dog/train/Cat/cat_10.jpg', 0), ('./cat_dog/train/Cat/cat_100.jpg', 0), ('./cat_dog/train/Cat/cat_1000.jpg', 0)]
dset_sizes:  {'train': 20000, 'val': 2000}

loader_train = torch.utils.data.DataLoader(dsets['train'], batch_size=64, shuffle=True, num_workers=6)
loader_valid = torch.utils.data.DataLoader(dsets['val'], batch_size=5, shuffle=False, num_workers=6)
loader_test = torch.utils.data.DataLoader(tsets['test'],batch_size=5,shuffle=False,num_workers=6)
'''
valid 数据一共有2000张图，每个batch是5张，因此，下面进行遍历一共会输出到 400
同时，把第一个 batch 保存到 inputs_try, labels_try，分别查看
'''
count = 1
for data in loader_valid:
    #print(count, end='\n')
    if count == 1:
        inputs_try,labels_try = data
    count +=1

print(labels_try)
print(inputs_try.shape)

tensor([0, 0, 0, 0, 0])
torch.Size([5, 3, 224, 224])

# 显示图片的小程序

def imshow(inp, title=None):
#   Imshow for Tensor.
    inp = inp.numpy().transpose((1, 2, 0))
    mean = np.array([0.485, 0.456, 0.406])
    std = np.array([0.229, 0.224, 0.225])
    inp = np.clip(std * inp + mean, 0,1)
    plt.imshow(inp)
    if title is not None:
        plt.title(title)
    plt.pause(0.001)  # pause a bit so that plots are updated
# 显示 labels_try 的5张图片，即valid里第一个batch的5张图片
out = torchvision.utils.make_grid(inputs_try)
imshow(out, title=[dset_classes[x] for x in labels_try])

创建VGG Model

torchvision中集成了很多在 ImageNet （120万张训练数据） 上预训练好的通用的CNN模型，可以直接下载使用。

在本课程中，我们直接使用预训练好的 VGG 模型。同时，为了展示 VGG 模型对本数据的预测结果，还下载了 ImageNet 1000 个类的 JSON 文件。

在这部分代码中，对输入的5个图片利用VGG模型进行预测，同时，使用softmax对结果进行处理，随后展示了识别结果。可以看到，识别结果是比较非常准确的。

!wget https://s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json

--2020-07-31 13:46:34--  https://s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json
Resolving s3.amazonaws.com (s3.amazonaws.com)... 54.231.49.164
Connecting to s3.amazonaws.com (s3.amazonaws.com)|54.231.49.164|:443... connected.
HTTP request sent, awaiting response... 200 OK
Length: 35363 (35K) [application/octet-stream]
Saving to: ‘imagenet_class_index.json’

imagenet_class_inde 100%[===================>]  34.53K  --.-KB/s    in 0.03s   

2020-07-31 13:46:34 (1.17 MB/s) - ‘imagenet_class_index.json’ saved [35363/35363]

model_vgg = models.vgg16(pretrained=True)

with open('./imagenet_class_index.json') as f:
    class_dict = json.load(f)
dic_imagenet = [class_dict[str(i)][1] for i in range(len(class_dict))]

inputs_try , labels_try = inputs_try.to(device), labels_try.to(device)
model_vgg = model_vgg.to(device)

outputs_try = model_vgg(inputs_try)

#print(outputs_try)
#print(outputs_try.shape)
#tensor([[-4.6803, -3.0721, -4.2074,  ..., -8.1783, -1.4379,  5.2827],
#        [-2.4916, -3.3212,  1.3284,  ..., -4.5295, -0.9055,  4.1661],
#        [-1.4204, -0.0192, -2.6073,  ..., -0.2028,  3.1158,  3.8306],
#        [-4.0369, -2.0386, -2.7258,  ..., -5.3328,  4.3880,  1.6959],
#        [-1.8230,  4.3508, -3.3690,  ..., -2.3910,  3.7018,  5.3185]],
#       device='cuda:0', grad_fn=)
#torch.Size([5, 1000])
'''
可以看到结果为5行，1000列的数据，每一列代表对每一种目标识别的结果。
但是我也可以观察到，结果非常奇葩，有负数，有正数，
为了将VGG网络输出的结果转化为对每一类的预测概率，我们把结果输入到 Softmax 函数
'''
m_softm = nn.Softmax(dim=1)
probs = m_softm(outputs_try)
vals_try,pred_try = torch.max(probs,dim=1)

#print( 'prob sum: ', torch.sum(probs,1))

#prob sum:  tensor([1.0000, 1.0000, 1.0000, 1.0000, 1.0000], device='cuda:0',
#      grad_fn=)

#print( 'vals_try: ', vals_try)

#vals_try:  tensor([0.9112, 0.2689, 0.4477, 0.5912, 0.4615], device='cuda:0',
#       grad_fn=)

#print( 'pred_try: ', pred_try)

#pred_try:  tensor([223, 223, 282, 285, 282], device='cuda:0')

print([dic_imagenet[i] for i in pred_try.data])
imshow(torchvision.utils.make_grid(inputs_try.data.cpu()), 
       title=[dset_classes[x] for x in labels_try.data.cpu()])

由此可见，VGG很强大可以识别猫的品种

修改最后一层，冻结前面层的参数

我们的目标是使用预训练好的模型，因此，需要把最后的 nn.Linear 层由1000类，替换为2类。为了在训练中冻结前面层的参数，需要设置 required_grad=False。这样，反向传播训练梯度时，前面层的权重就不会自动更新了。训练中，只会更新最后一层的参数。

print(model_vgg)

model_vgg_new = model_vgg;

for param in model_vgg_new.parameters():
    param.requires_grad = False
model_vgg_new.classifier._modules['6'] = nn.Linear(4096, 2)
model_vgg_new.classifier._modules['7'] = torch.nn.LogSoftmax(dim = 1)

model_vgg_new = model_vgg_new.to(device)

print(model_vgg_new.classifier)

VGG(
  (features): Sequential(
    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU(inplace=True)
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (6): ReLU(inplace=True)
    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (8): ReLU(inplace=True)
    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (11): ReLU(inplace=True)
    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (13): ReLU(inplace=True)
    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (15): ReLU(inplace=True)
    (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (18): ReLU(inplace=True)
    (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (20): ReLU(inplace=True)
    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (22): ReLU(inplace=True)
    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (25): ReLU(inplace=True)
    (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (27): ReLU(inplace=True)
    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (29): ReLU(inplace=True)
    (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
  (classifier): Sequential(
    (0): Linear(in_features=25088, out_features=4096, bias=True)
    (1): ReLU(inplace=True)
    (2): Dropout(p=0.5, inplace=False)
    (3): Linear(in_features=4096, out_features=4096, bias=True)
    (4): ReLU(inplace=True)
    (5): Dropout(p=0.5, inplace=False)
    
    
    (6): Linear(in_features=4096, out_features=2, bias=True)
    
    
    
    (7): LogSoftmax()
  )
)
Sequential(
  (0): Linear(in_features=25088, out_features=4096, bias=True)
  (1): ReLU(inplace=True)
  (2): Dropout(p=0.5, inplace=False)
  (3): Linear(in_features=4096, out_features=4096, bias=True)
  (4): ReLU(inplace=True)
  (5): Dropout(p=0.5, inplace=False)
  
  
  
  (6): Linear(in_features=4096, out_features=2, bias=True)
  
  
  
  (7): LogSoftmax()
)

训练并测试全连接层

Adam更加准确

'''
第一步：创建损失函数和优化器

损失函数 NLLLoss() 的 输入 是一个对数概率向量和一个目标标签. 
它不会为我们计算对数概率，适合最后一层是log_softmax()的网络. 
'''
criterion = nn.NLLLoss()

# 学习率
lr = 0.001

# 随机梯度下降
#optimizer_vgg = torch.optim.SGD(model_vgg_new.classifier[6].parameters(),lr = lr)
optimizer_vgg = torch.optim.Adam(model_vgg_new.classifier[6].parameters(),lr = lr)

'''
第二步：训练模型
'''

def train_model(model,dataloader,size,epochs=1,optimizer=None):
    model.train()
    
    for epoch in range(epochs):
        running_loss = 0.0
        running_corrects = 0
        count = 0
        for inputs,classes in dataloader:
            inputs = inputs.to(device)
            classes = classes.to(device)
            outputs = model(inputs)
            loss = criterion(outputs,classes)           
            optimizer = optimizer
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
            _,preds = torch.max(outputs.data,1)
            # statistics
            running_loss += loss.data.item()
            running_corrects += torch.sum(preds == classes.data)
            count += len(inputs)
            print('Training: No. ', count, ' process ... total: ', size)
        epoch_loss = running_loss / size
        epoch_acc = running_corrects.data.item() / size
        print('Loss: {:.4f} Acc: {:.4f}'.format(
                     epoch_loss, epoch_acc))
        
        
# 模型训练
train_model(model_vgg_new,loader_train,size=dset_sizes['train'], epochs=1, 
            optimizer=optimizer_vgg)

Loss: 0.0024 Acc: 0.9518

def test_model(model,dataloader,size):
    model.eval()
    predictions = np.zeros(size)
    all_classes = np.zeros(size)
    all_proba = np.zeros((size,2))
    i = 0
    running_loss = 0.0
    running_corrects = 0
    for inputs,classes in dataloader:
        inputs = inputs.to(device)
        classes = classes.to(device)
        outputs = model(inputs)
        loss = criterion(outputs,classes)           
        _,preds = torch.max(outputs.data,1)
        # statistics
        running_loss += loss.data.item()
        running_corrects += torch.sum(preds == classes.data)
        predictions[i:i+len(classes)] = preds.to('cpu').numpy()
        all_classes[i:i+len(classes)] = classes.to('cpu').numpy()
        all_proba[i:i+len(classes),:] = outputs.data.to('cpu').numpy()
        i += len(classes)
        print('Testing: No. ', i, ' process ... total: ', size)        
    epoch_loss = running_loss / size
    epoch_acc = running_corrects.data.item() / size
    print('Loss: {:.4f} Acc: {:.4f}'.format(
                     epoch_loss, epoch_acc))
    return predictions, all_proba, all_classes

# 测试网络（valid）
predictions, all_proba, all_classes = test_model(model_vgg_new,loader_valid,size=dset_sizes['val'])

Loss: 0.0179 Acc: 0.9735