● 本文为365天深度学习训练营中的学习记录博客● 原作者K同学啊学习目标1.编写一个完整的深度学习程序2. 手动推导卷积层与池化层的计算过程一、前期准备1.设置GPUimport torch import torch.nn as nn import matplotlib.pyplot as plt import torchvision device torch.device(cuda if torch.cuda.is_available() else cpu) device2.导入数据使用dataset下载CIFAR10数据并划分训练集和测试集train_ds torchvision.datasets.CIFAR10(data, train True, transform torchvision.transforms.ToTensor(), download True) test_ds torchvision.datasets.CIFAR10(data, train False, transform torchvision.transforms.ToTensor(), download True)使用DataLoader加载数据并取出一批查看其输出格式batch_size 32 train_dl torch.utils.data.DataLoader(train_ds, batch_size batch_size, shuffle True) #打乱 test_dl torch.utils.data.DataLoader(test_ds, batch_size batch_size) imgs,labels next(iter(train_dl)) #取出一批的图像和对应标签 imgs.shape #输出格式[batch_size,channel,height,width]3.数据可视化transpose((1,2,0))把轴的顺序由[C,H,W]变为[H,W,C]更便于数据处理import numpy as np # 指定图片大小图像大小为20宽、5高的绘图(单位为英寸inch) plt.figure(figsize(20, 5)) for i, imgs in enumerate(imgs[:20]): # 进行轴变换 npimg imgs.numpy().transpose((1, 2, 0)) # 将整个figure分成2行10列绘制第i1个子图。 plt.subplot(2, 10, i1) plt.imshow(npimg, cmapplt.cm.binary) plt.axis(off) plt.show()二、构建CNN网络1.相关函数详解torch.nn.Cov2d()torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride1, padding0, dilation1, groups1, biasTrue, padding_modezeros, deviceNone, dtypeNone)in_channels(int)输入图像的通道数out_channels(int)卷积产生的通道数kernel_size(int / tuple)卷积核的大小stride(int / tuple)卷积的步幅默认为1padding(int / tuple / str)添加到输入的所有四个边的填充默认为0dilation(int / tuple)扩张操作卷积核点的距离默认为1groups(int)将输入通道分组成多个子组每个子组使用一组卷积核来处理。默认值为1表示不进行分组卷积。padding(str)zeros, reflect, replicate或circular. 默认zerostorch.nn.Linear()torch.nn.Linear(in_features, out_features, biasTrue, deviceNone, dtypeNone)in_features每个输入样本的大小out_features每个输出样本的大小torch.nn.MaxPool2d()torch.nn.MaxPool2d(kernel_size, strideNone, padding0, dilation1, return_indicesFalse, ceil_modeFalse)kernel_size最大的窗口大小stride窗口的步幅默认值为kernel_sizepadding填充值默认为0dilation控制窗口中元素步幅的参数2.构建模型import torch.nn.functional as F num_class 10 class Model(nn.Module): def __init__(self): super().__init__() #特征提取网络 self.conv1 nn.Conv2d(3,64,kernel_size3) self.pool1 nn.MaxPool2d(kernel_size2) self.conv2 nn.Conv2d(64,64,kernel_size3) self.pool2 nn.MaxPool2d(kernel_size2) self.conv3 nn.Conv2d(64,128,kernel_size3) self.pool3 nn.MaxPool2d(kernel_size2) #分类网络 self.fc1 nn.Linear(512,256) self.fc2 nn.Linear(256,num_class) def forward(self,x): x self.pool1(F.relu(self.conv1(x))) x self.pool2(F.relu(self.conv2(x))) x self.pool3(F.relu(self.conv3(x))) x torch.flatten(x,start_dim1) x F.relu(self.fc1(x)) x self.fc2(x) return x加载并打印模型from torchinfo import summary model Model().to(device) summary(model)3.卷积层和池化层的计算上图为池化层的计算公式下图为我的推导过程三、训练模型1.设置超参数loss_fn nn.CrossEntropyLoss() #损失函数 learn_rate 1e-2 #学习率 opt torch.optim.SGD(model.parameters(),lr learn_rate)2.编写训练函数optimizer.zero_grad()清空模型的梯度缓存loss.backward()反向传播自动计算每个参数的梯度optimizer.step()根据当前计算的梯度更新模型的参数def train(dataloader,model,loss_fn,optimizer): size len(dataloader.dataset) num_batches len(dataloader) train_loss,train_acc 0,0 for X,y in dataloader: X,y X.to(device),y.to(device) pred model(X) loss loss_fn(pred,y) optimizer.zero_grad() loss.backward() optimizer.step() train_acc (pred.argmax(1) y).type(torch.float).sum().item() train_loss loss.item() train_acc / size train_loss / num_batches return train_acc,train_loss3.编写测试函数def test(dataloader,model,loss_fn): size len(dataloader.dataset) num_batches len(dataloader) test_loss,test_acc 0,0 with torch.no_grad(): for imgs,target in dataloader: imgs,target imgs.to(device),target.to(device) target_pred model(imgs) loss loss_fn(target_pred,target) test_loss loss.item() test_acc (target_pred.argmax(1) target).type(torch.float).sum().item() test_acc / size test_loss / num_batches return test_acc,test_loss4.正式训练model.train()开始训练model.eval()开始评估epochs 10 train_loss [] train_acc [] test_loss [] test_acc [] for epoch in range(epochs): model.train() epoch_train_acc,epoch_train_loss train(train_dl,model,loss_fn,opt) model.eval() epoch_test_acc,epoch_test_loss test(test_dl,model,loss_fn) train_acc.append(epoch_train_acc) train_loss.append(epoch_train_loss) test_acc.append(epoch_test_acc) test_loss.append(epoch_test_loss) template (Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%Test_loss:{:.3f}) print(template.format(epoch1, epoch_train_acc*100, epoch_train_loss, epoch_test_acc*100, epoch_test_loss)) print(Done)四、结果可视化import matplotlib.pyplot as plt #隐藏警告 import warnings warnings.filterwarnings(ignore) #忽略警告信息 plt.rcParams[font.sans-serif] [SimHei] # 用来正常显示中文标签 plt.rcParams[axes.unicode_minus] False # 用来正常显示负号 plt.rcParams[figure.dpi] 100 #分辨率 from datetime import datetime current_time datetime.now() # 获取当前时间 epochs_range range(epochs) plt.figure(figsize(12, 3)) plt.subplot(1, 2, 1) plt.plot(epochs_range, train_acc, labelTraining Accuracy) plt.plot(epochs_range, test_acc, labelTest Accuracy) plt.legend(loclower right) plt.title(Training and Validation Accuracy) plt.xlabel(current_time) plt.subplot(1, 2, 2) plt.plot(epochs_range, train_loss, labelTraining Loss) plt.plot(epochs_range, test_loss, labelTest Loss) plt.legend(locupper right) plt.title(Training and Validation Loss) plt.show()五、个人总结这周进一步学习了CNN网络了解了卷积层、池化层、全连接层的函数并学习了卷积层和池化层的计算。与P1周使用的MINIST灰色图像不同本周使用的数据集是CIFAR10是彩色图像在可视化时需要使用transpose()进行轴变换这里建立的CNN网络是输入层-卷积层1-池化层1-卷积层2-池化层2-卷积层3-池化层3-flatten层-全连接层-输出层比P1建立的CNN网络复杂一些并且进行了10次训练但最终准确率远低于手写数字识别也可以看出彩色图像是更加难以识别的。