【lenet 图片分类】

# -*- coding: utf-8 -*-
"""
Created on Tue Dec 13 19:15:49 2022
"""

import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torchvision.transforms as transforms
# optim中定义了各种各样的优化方法，包括SGD
import torch.optim as optim
import datetime
 
 
class LeNet(nn.Module):
    # 一般在__init__中定义网络需要的操作算子，比如卷积、全连接算子等等
    def __init__(self):
        super(LeNet, self).__init__()
        # Conv2d的第一个参数是输入的channel数量，第二个是输出的channel数量，第三个是kernel size
        self.conv1 = nn.Conv2d(3, 6, 5)
        self.conv2 = nn.Conv2d(6, 16, 5)
        # 由于上一层有16个channel输出，每个feature map大小为5*5，所以全连接层的输入是16*5*5
        self.fc1 = nn.Linear(16*5*5, 120)
        self.fc2 = nn.Linear(120, 84)
        # 最终有10类，所以最后一个全连接层输出数量是10
        self.fc3 = nn.Linear(84, 10)
        self.pool = nn.MaxPool2d(2, 2)
    # forward这个函数定义了前向传播的运算，只需要像写普通的python算数运算那样就可以了
    def forward(self, x):
        # 输入x:torch.Size([32, 3, 32, 32])
        # self.conv1(x), Conv2d(3, 6, 5): torch.Size([32, 6, 28, 28])
        # F.relu输出x: torch.Size([32, 6, 28, 28])
        x = F.relu(self.conv1(x))
        # self.pool(x)：torch.Size([32, 6, 14, 14])
        x = self.pool(x)
        # self.conv2(x)：torch.Size([32, 16, 10, 10])
        # F.relu输出x：torch.Size([32, 16, 10, 10])
        x = F.relu(self.conv2(x))
        # self.pool(x)：torch.Size([32, 16, 5, 5])
        x = self.pool(x)
        # 下面这步把二维特征图变为一维，这样全连接层才能处理
        # x.view(-1, 16*5*5)：torch.Size([32, 400])
        x = x.view(-1, 16*5*5)
        # self.fc1(x)：torch.Size([32, 120])
        # F.relu：torch.Size([32, 120])
        x = F.relu(self.fc1(x))
        # self.fc1(x)：torch.Size([32, 84])
        # F.relu：torch.Size([32, 84])
        x = F.relu(self.fc2(x))
        # self.fc3：torch.Size([32, 10])
        x = self.fc3(x)
        return x
# cifar-10官方提供的数据集是用numpy array存储的
# 下面这个transform会把numpy array变成torch tensor，然后把rgb值归一到[0, 1]这个区间
transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
 
# 在构建数据集的时候指定transform，就会应用我们定义好的transform
# root是存储数据的文件夹，download=True指定如果数据不存在先下载数据
cifar_train = torchvision.datasets.CIFAR10(root='./data', train=True,
                                           download=True, transform=transform)
cifar_test = torchvision.datasets.CIFAR10(root='./data', train=False,
                                          transform=transform)
 
print(cifar_train)
print(cifar_test)
trainloader = torch.utils.data.DataLoader(cifar_train, batch_size=32, shuffle=True)
testloader = torch.utils.data.DataLoader(cifar_test, batch_size=32, shuffle=True)
 
 
# 由于需要用到GPU，所以先获取device，然后再把网络的参数复制到GPU上
# 如果你没有GPU，那么可以忽略device相关的代码
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
net = LeNet().to(device)
 
# CrossEntropyLoss就是我们需要的损失函数
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
 
if __name__ == '__main__':
    print("Start Training...")
    start_time = datetime.datetime.now()
    print("start_time:", start_time)
    for epoch in range(5):
        # 我们用一个变量来记录每100个batch的平均loss
        loss100 = 0.0
        # 我们的dataloader派上了用场
        for i, data in enumerate(trainloader):
            inputs, labels = data
            inputs, labels = inputs.to(device), labels.to(device)  # 注意需要复制到GPU
            optimizer.zero_grad()
            outputs = net(inputs)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()
            loss100 += loss.item()
            if i % 100 == 99:
                print('[Epoch %d, Batch %5d] loss: %.3f' %
                      (epoch + 1, i + 1, loss100 / 100))
                loss100 = 0.0
 
    end_time = datetime.datetime.now()
    print("end_time:", end_time)
    print("Done Training!")
 
    delta = end_time - start_time
    # 获取timedelta对象包含的总秒数
    print("total_seconds: ", delta.total_seconds())
    
    torch.save(net, './data/model.pth')  # 将模型保存到当前目录下 data 文件夹内，名为 model.pth
    # 构造测试的dataloader
dataiter = iter(testloader)
# 预测正确的数量和总数量
#%%
import numpy
correct = 0
total = 0
# 使用torch.no_grad的话在前向传播中不记录梯度，节省内存
exp=[]
pred=[]
with torch.no_grad():
    for data in testloader:
        images, labels = data
        images, labels = images.to(device), labels.to(device)
        # 预测
        outputs = net(images)
        # 我们的网络输出的实际上是个概率分布，去最大概率的哪一项作为预测分类
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()
        exp=numpy.append(exp,labels.numpy())
        pred=numpy.append(pred,predicted.numpy())

print('Accuracy of the network on the 10000 test images: %d %%' % (
    100 * correct / total))
#%%
print(exp)
print(pred)
from sklearn import metrics
EMOS=['plane','car','bird','cat','deer','dog','frog','horse','ship','cart']

print(metrics.classification_report(exp, pred,target_names=EMOS, digits=4, labels=list(range(10))))