pytorch构建卷积神经网络流程

• 卷积网络中的输入和输出层与传统神经网络有些区别，需重新设计，训练模块基本一致

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

首先读取数据

• 分别构建训练集和测试集（验证集）
• DataLoader来迭代读取数据

# 定义超参数 
input_size = 28  #图像的总尺寸28*28
num_classes = 10  #标签的种类数
num_epochs = 3  #训练的总循环周期
batch_size = 64  #一个撮（批次）的大小，64张图片

# 训练集
train_dataset = datasets.MNIST(root='./data',  
                            train=True,   
                            transform=transforms.ToTensor(), #把一个取值范围是[0,255]的PIL.Image或者shape为(H,W,C)的numpy.ndarray，转换成形状为[C,H,W]，取值范围是[0,1.0]的torch.FloadTensor 
                            download=True) 

# 测试集
test_dataset = datasets.MNIST(root='./data', 
                           train=False, 
                           transform=transforms.ToTensor())

# 构建batch数据
train_loader = torch.utils.data.DataLoader(dataset=train_dataset, 
                                           batch_size=batch_size, 
                                           shuffle=True)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset, 
                                           batch_size=batch_size, 
                                           shuffle=True)

卷积网络模块构建

• 一般卷积层，relu层，池化层可以写成一个套餐
• 注意卷积最后结果还是一个特征图，需要把图转换成向量才能做分类或者回归任务

class CNN(nn.Module):
    def __init__(self):
        super(CNN, self).__init__()
        self.conv1 = nn.Sequential(         # 输入大小 (1, 28, 28)
            nn.Conv2d(
                in_channels=1,              # 灰度图  输入特征图
                out_channels=16,            # 要得到几多少个特征图 用多少个卷积和进行提取
                kernel_size=5,              # 卷积核大小
                stride=1,                   # 步长
                padding=2,                  # 如果希望卷积后大小跟原来一样，需要设置padding=(kernel_size-1)/2 if stride=1
            ),                              # 输出的特征图为 (16, 28, 28)
            nn.ReLU(),                      # relu层
            nn.MaxPool2d(kernel_size=2),    # 进行池化操作（2x2 区域）, 输出结果为： (16, 14, 14)
        )
        self.conv2 = nn.Sequential(         # 下一个套餐的输入 (16, 14, 14)
            nn.Conv2d(16, 32, 5, 1, 2),     # 输出 (32, 14, 14)
            nn.ReLU(),                      # relu层
            nn.MaxPool2d(2),                # 输出 (32, 7, 7)
        )
        self.out = nn.Linear(32 * 7 * 7, 10)   # 全连接层得到的结果

    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = x.view(x.size(0), -1)           # flatten操作，结果为：(batch_size, 32 * 7 * 7)  把图像拉成一维向量做全连接操作
        output = self.out(x)   #全连接操作
        return output

准确率作为评估标准

def accuracy(predictions, labels):
    pred = torch.max(predictions.data, 1)[1] 
    rights = pred.eq(labels.data.view_as(pred)).sum() 
    return rights, len(labels) 

训练网络模型

# 实例化
net = CNN() 
#损失函数
criterion = nn.CrossEntropyLoss() 
#优化器
optimizer = optim.Adam(net.parameters(), lr=0.001) #定义优化器，普通的随机梯度下降算法

#开始训练循环
for epoch in range(num_epochs):
    #当前epoch的结果保存下来
    train_rights = [] 
    
    for batch_idx, (data, target) in enumerate(train_loader):  #针对容器中的每一个批进行循环
        net.train()                             
        output = net(data) 
        loss = criterion(output, target) 
        optimizer.zero_grad() 
        loss.backward() 
        optimizer.step() 
        right = accuracy(output, target) 
        train_rights.append(right) 

    
        if batch_idx % 100 == 0: 
            
            net.eval() 
            val_rights = [] 
            
            for (data, target) in test_loader:
                output = net(data) 
                right = accuracy(output, target) 
                val_rights.append(right)
                
            #准确率计算
            train_r = (sum([tup[0] for tup in train_rights]), sum([tup[1] for tup in train_rights]))
            val_r = (sum([tup[0] for tup in val_rights]), sum([tup[1] for tup in val_rights]))

            print('当前epoch: {} [{}/{} ({:.0f}%)]\t损失: {:.6f}\t训练集准确率: {:.2f}%\t测试集正确率: {:.2f}%'.format(
                epoch, batch_idx * batch_size, len(train_loader.dataset),
                100. * batch_idx / len(train_loader), 
                loss.data, 
                100. * train_r[0].numpy() / train_r[1], 
                100. * val_r[0].numpy() / val_r[1]))

当前epoch: 0 [0/60000 (0%)] 损失: 2.304030 训练集准确率: 6.25% 测试集正确率: 12.26%
当前epoch: 0 [6400/60000 (11%)] 损失: 0.250382 训练集准确率: 77.20% 测试集正确率: 92.99%
当前epoch: 0 [12800/60000 (21%)] 损失: 0.067399 训练集准确率: 85.38% 测试集正确率: 94.79%
当前epoch: 0 [19200/60000 (32%)] 损失: 0.201983 训练集准确率: 88.58% 测试集正确率: 96.09%
当前epoch: 0 [25600/60000 (43%)] 损失: 0.218886 训练集准确率: 90.61% 测试集正确率: 96.83%
当前epoch: 0 [32000/60000 (53%)] 损失: 0.028800 训练集准确率: 91.81% 测试集正确率: 97.59%
当前epoch: 0 [38400/60000 (64%)] 损失: 0.244649 训练集准确率: 92.63% 测试集正确率: 97.50%
当前epoch: 0 [44800/60000 (75%)] 损失: 0.034414 训练集准确率: 93.33% 测试集正确率: 97.96%
当前epoch: 0 [51200/60000 (85%)] 损失: 0.046361 训练集准确率: 93.83% 测试集正确率: 98.11%
当前epoch: 0 [57600/60000 (96%)] 损失: 0.052881 训练集准确率: 94.25% 测试集正确率: 97.61%
当前epoch: 1 [0/60000 (0%)] 损失: 0.014872 训练集准确率: 100.00% 测试集正确率: 98.01%
当前epoch: 1 [6400/60000 (11%)] 损失: 0.037074 训练集准确率: 97.82% 测试集正确率: 98.19%
当前epoch: 1 [12800/60000 (21%)] 损失: 0.067327 训练集准确率: 98.03% 测试集正确率: 98.27%
当前epoch: 1 [19200/60000 (32%)] 损失: 0.009921 训练集准确率: 98.11% 测试集正确率: 98.18%
当前epoch: 1 [25600/60000 (43%)] 损失: 0.060707 训练集准确率: 98.13% 测试集正确率: 98.52%
当前epoch: 1 [32000/60000 (53%)] 损失: 0.036018 训练集准确率: 98.20% 测试集正确率: 98.54%
当前epoch: 1 [38400/60000 (64%)] 损失: 0.162394 训练集准确率: 98.25% 测试集正确率: 98.64%
当前epoch: 1 [44800/60000 (75%)] 损失: 0.016710 训练集准确率: 98.27% 测试集正确率: 98.51%
当前epoch: 1 [51200/60000 (85%)] 损失: 0.015306 训练集准确率: 98.27% 测试集正确率: 98.43%
当前epoch: 1 [57600/60000 (96%)] 损失: 0.032269 训练集准确率: 98.28% 测试集正确率: 98.63%
当前epoch: 2 [0/60000 (0%)] 损失: 0.026483 训练集准确率: 100.00% 测试集正确率: 98.90%
当前epoch: 2 [6400/60000 (11%)] 损失: 0.059850 训练集准确率: 98.62% 测试集正确率: 98.72%
当前epoch: 2 [12800/60000 (21%)] 损失: 0.132063 训练集准确率: 98.62% 测试集正确率: 98.48%
当前epoch: 2 [19200/60000 (32%)] 损失: 0.032505 训练集准确率: 98.65% 测试集正确率: 98.77%
当前epoch: 2 [25600/60000 (43%)] 损失: 0.022765 训练集准确率: 98.67% 测试集正确率: 98.81%
当前epoch: 2 [32000/60000 (53%)] 损失: 0.018788 训练集准确率: 98.70% 测试集正确率: 98.70%
当前epoch: 2 [38400/60000 (64%)] 损失: 0.075944 训练集准确率: 98.70% 测试集正确率: 99.01%
当前epoch: 2 [44800/60000 (75%)] 损失: 0.013405 训练集准确率: 98.74% 测试集正确率: 98.75%
当前epoch: 2 [51200/60000 (85%)] 损失: 0.015860 训练集准确率: 98.72% 测试集正确率: 99.00%
当前epoch: 2 [57600/60000 (96%)] 损失: 0.001865 训练集准确率: 98.73% 测试集正确率: 98.93%

数据集参考