第四周-P4猴痘识别
>- ** 本文为[365天深度学习训练营](https://mp.weixin.qq.com/s/Nb93582M_5usednAKp_Jtw) 中的学习记录博客**
>- ** 参考文章：[Pytorch实战 | 第P4周：猴痘病识别](https://www.heywhale.com/mw/project/6347b0065565973b87564268)**
>- ** 原作者：[K同学啊 | 接辅导、项目定制](https://mtyjkh.blog.csdn.net/)**
>- ** 文章来源：[K同学的学习圈子](https://www.yuque.com/mingtian-fkmxf/zxwb45)**
要求：
- 训练过程中保存效果最好的模型参数。
- 加载最佳模型参数识别本地的一张图片。
- 调整网络结构使测试集accuracy到达88%（重点）。（完成）
拔高（可选）：
- 调整模型参数并观察测试集的准确率变化。
- 尝试设置动态学习率。
- 测试集accuracy到达90%。
一、前期准备
1.设置gpu
#加载需要的包
import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms,datasets
import os,PIL,pathlib,random
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
device2.加载数据
#第一步使用pathlib.Path()将字符串类型文件路径转换为pathlib.Path对象
#第二步使用glob()方法获取data_dir目录下其他子文件路径
#第三步使用split函数将字符串路径分割，以此获取各个文件所属的类别名称
import os,PIL,random,pathlib
data_dir = r"F:\P4_data"
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob("*"))
classNames = [str(path).split("\\")[2] for path in data_paths]
data_Paths
classNames[WindowsPath('F:/P4_data/Monkeypox'), WindowsPath('F:/P4_data/Others')]['Monkeypox', 'Others']
3.图片可视化（猴痘图片可能会有些不适，不予展示）
import matplotlib.pyplot as plt
from PIL import Image
#获取图像的文件夹路径
image_folder = r"F:\P4_data\Monkeypox"
#获取各个图片路径后缀
image_files = [f for f in os.listdir(image_folder) if f.endswith((".jpg",".png",".jpeg")) ]
#创建Matplotlib图像
fig, axes = plt.subplots(3, 8, figsize=(16, 6))
for ax, img_file in zip(axes.flat, image_files):
img_path = os.path.join(image_folder,img_file)
img = Image.open(img_path)
ax.imshow(img)
ax.axis("off")
4.转换数据
total_datadir = r"F:\P4_data"
train_transforms = transforms.Compose([
transforms.Resize([224,224]),
transforms.ToTensor(),
transforms.Normalize(
mean=[0.485,0.456,0.406],
std=[0.229,0.224,0.225])
])
total_data = datasets.ImageFolder(total_datadir,transform=train_transforms)
total_data
Dataset ImageFolder
Number of datapoints: 2142
Root location: F:\P4_data
StandardTransform
Transform: Compose(
Resize(size=[224, 224], interpolation=bilinear, max_size=None, antialias=warn)
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
total_data.class_to_idx
{'Monkeypox': 0, 'Others': 1}
5.划分训练集和测试集
train_size = int(0.8 * len(total_data))
test_size = len(total_data) - train_size
train_dataset, test_dataset = torch.utils.data.random_split(total_data,[train_size,test_size])
train_dataset, test_dataset
train_size,test_size(, ) (1713, 429)
6.加载数据集
batch_size = 32
train_dl = torch.utils.data.DataLoader(train_dataset,
batch_size = batch_size,
shuffle=True,
num_workers=1)
test_dl = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=1)
for X,y in test_dl:
print("Shape of X [N,C,H,W]:",X.shape)
print("Shape of y:",y.shape,y.dtype)
breakShape of X [N,C,H,W]: torch.Size([32, 3, 224, 224]) Shape of y: torch.Size([32]) torch.int64
二、构建简单的CNN神经网络
构建了卷积-池化的CNN模型
import torch.nn.functional as F
class Network_bn(nn.Module):
def __init__(self):
super(Network_bn, self).__init__()
self.conv1 = nn.Conv2d(in_channels=3, out_channels=12, kernel_size=5, stride=1, padding=0)
self.bn1 = nn.BatchNorm2d(12)
self.conv2 = nn.Conv2d(in_channels=12,out_channels=12, kernel_size=5, stride=1, padding=0)
self.bn2 =nn.BatchNorm2d(12)
self.pool = nn.MaxPool2d(2,2)
self.conv4 = nn.Conv2d(in_channels=12,out_channels=24, kernel_size=5, stride=1, padding=0)
self.bn4 = nn.BatchNorm2d(24)
self.conv5 = nn.Conv2d(in_channels=24,out_channels=24, kernel_size=5, stride=1, padding =0)
self.bn5 = nn.BatchNorm2d(24)
self.conv7 = nn.Conv2d(in_channels=24,out_channels=48, kernel_size=5, stride=1, padding=0)
self.bn7 = nn.BatchNorm2d(48)
self.conv8 = nn.Conv2d(in_channels=48,out_channels=48, kernel_size=5, stride=1, padding=0)
self.bn8 = nn.BatchNorm2d(48)
self.fc1 = nn.Linear(48*21*21,len(classNames))
def forward(self,x ):
x = F.relu(self.bn1(self.conv1(x)))
x = F.relu(self.bn2(self.conv2(x)))
x = self.pool(x)
x = F.relu(self.bn4(self.conv4(x)))
x = F.relu(self.bn5(self.conv5(x)))
x = self.pool(x)
x = F.relu(self.bn7(self.conv7(x)))
x = F.relu(self.bn8(self.conv8(x)))
x = self.pool(x)
x = x.view(-1, 48*21*21)
x = self.fc1(x)
return x
device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))
model = Network_bn().to(device)
model三、训练模型
1.设置超参数
loss_fn = nn.CrossEntropyLoss()
learn_rate = 0.001
opt = torch.optim.SGD(model.parameters(),lr=learn_rat2.编写训练函数
# 训练循环
def train(dataloader, model, loss_fn, optimizer):
size = len(dataloader.dataset)
num_batches = len(dataloader)
train_loss ,train_acc = 0,0 # 初始化训练损失和正确率
for X,y in dataloader:
X,y = X.to(device),y.to(device)
#计算预测误差
pred = model(X)
loss =loss_fn(pred,y)
#反向传播
optimizer.zero_grad() #grad属性归零
loss.backward()
optimizer.step()
#记录acc和loss
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_loss 3.编写测试函数
##测试函数
def test(dataloader, model, loss_fn):
size = len(dataloader.dataset)
num_batches = len(dataloader)
test_loss, test_acc = 0 , 0
#当不进行训练时，停止梯度更新，节省计算内存消耗：
with torch.no_grad():
for imgs, target in dataloader:
imgs, target = imgs.to(device), target.to(device)
#计算loss
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
epochs = 50
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) # Correct the variable name here, it should be `epoch_test_loss`, not `epoch_test_acc`
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:{}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}' # Fixed the format specifier for epoch
print(template.format(epoch + 1, epoch_train_acc * 100, epoch_train_loss, epoch_test_acc * 100, epoch_test_loss))
print("Done")Epoch:1, Train_acc:61.1%, Train_loss:0.891, Test_acc:69.0%, Test_loss:0.569 Epoch:2, Train_acc:70.0%, Train_loss:0.616, Test_acc:56.6%, Test_loss:0.704 Epoch:3, Train_acc:73.4%, Train_loss:0.566, Test_acc:72.3%, Test_loss:0.617 Epoch:4, Train_acc:81.1%, Train_loss:0.421, Test_acc:79.7%, Test_loss:0.428 Epoch:5, Train_acc:86.2%, Train_loss:0.342, Test_acc:79.5%, Test_loss:0.432 Epoch:6, Train_acc:88.4%, Train_loss:0.300, Test_acc:76.9%, Test_loss:0.523 Epoch:7, Train_acc:88.4%, Train_loss:0.288, Test_acc:73.2%, Test_loss:0.620 Epoch:8, Train_acc:90.4%, Train_loss:0.254, Test_acc:85.3%, Test_loss:0.352 Epoch:9, Train_acc:92.1%, Train_loss:0.226, Test_acc:85.1%, Test_loss:0.339 Epoch:10, Train_acc:92.1%, Train_loss:0.211, Test_acc:87.6%, Test_loss:0.313 Epoch:11, Train_acc:94.5%, Train_loss:0.178, Test_acc:86.7%, Test_loss:0.352 Epoch:12, Train_acc:94.3%, Train_loss:0.166, Test_acc:84.6%, Test_loss:0.324 Epoch:13, Train_acc:94.6%, Train_loss:0.156, Test_acc:87.4%, Test_loss:0.307 Epoch:14, Train_acc:96.0%, Train_loss:0.143, Test_acc:86.7%, Test_loss:0.279 Epoch:15, Train_acc:96.2%, Train_loss:0.134, Test_acc:83.4%, Test_loss:0.347 Epoch:16, Train_acc:96.7%, Train_loss:0.129, Test_acc:88.3%, Test_loss:0.305 Epoch:17, Train_acc:97.3%, Train_loss:0.116, Test_acc:86.7%, Test_loss:0.284 Epoch:18, Train_acc:97.5%, Train_loss:0.108, Test_acc:87.4%, Test_loss:0.298 Epoch:19, Train_acc:97.5%, Train_loss:0.099, Test_acc:86.2%, Test_loss:0.314 Epoch:20, Train_acc:97.2%, Train_loss:0.099, Test_acc:88.8%, Test_loss:0.274 Epoch:21, Train_acc:97.7%, Train_loss:0.094, Test_acc:87.9%, Test_loss:0.273 Epoch:22, Train_acc:98.6%, Train_loss:0.080, Test_acc:87.4%, Test_loss:0.278 Epoch:23, Train_acc:98.0%, Train_loss:0.091, Test_acc:88.1%, Test_loss:0.282 Epoch:24, Train_acc:98.7%, Train_loss:0.069, Test_acc:87.6%, Test_loss:0.309 Epoch:25, Train_acc:98.7%, Train_loss:0.073, Test_acc:88.3%, Test_loss:0.315 Epoch:26, Train_acc:98.8%, Train_loss:0.068, Test_acc:88.8%, Test_loss:0.271 Epoch:27, Train_acc:98.9%, Train_loss:0.060, Test_acc:89.3%, Test_loss:0.301 Epoch:28, Train_acc:98.8%, Train_loss:0.069, Test_acc:86.9%, Test_loss:0.288 Epoch:29, Train_acc:99.2%, Train_loss:0.054, Test_acc:88.6%, Test_loss:0.295 Epoch:30, Train_acc:99.3%, Train_loss:0.051, Test_acc:87.6%, Test_loss:0.372 Epoch:31, Train_acc:99.4%, Train_loss:0.052, Test_acc:88.1%, Test_loss:0.293 Epoch:32, Train_acc:99.5%, Train_loss:0.046, Test_acc:89.5%, Test_loss:0.282 Epoch:33, Train_acc:99.6%, Train_loss:0.043, Test_acc:89.0%, Test_loss:0.265 Epoch:34, Train_acc:99.2%, Train_loss:0.047, Test_acc:89.5%, Test_loss:0.274 Epoch:35, Train_acc:99.6%, Train_loss:0.039, Test_acc:88.3%, Test_loss:0.289 Epoch:36, Train_acc:99.6%, Train_loss:0.040, Test_acc:89.3%, Test_loss:0.304 Epoch:37, Train_acc:99.6%, Train_loss:0.035, Test_acc:89.7%, Test_loss:0.284 Epoch:38, Train_acc:99.4%, Train_loss:0.041, Test_acc:88.8%, Test_loss:0.277 Epoch:39, Train_acc:99.6%, Train_loss:0.037, Test_acc:88.8%, Test_loss:0.262 Epoch:40, Train_acc:99.7%, Train_loss:0.037, Test_acc:89.0%, Test_loss:0.284 Epoch:41, Train_acc:99.7%, Train_loss:0.033, Test_acc:88.8%, Test_loss:0.283 Epoch:42, Train_acc:99.5%, Train_loss:0.036, Test_acc:87.6%, Test_loss:0.296 Epoch:43, Train_acc:99.9%, Train_loss:0.027, Test_acc:89.7%, Test_loss:0.266 Epoch:44, Train_acc:99.8%, Train_loss:0.028, Test_acc:89.3%, Test_loss:0.285 Epoch:45, Train_acc:99.7%, Train_loss:0.029, Test_acc:88.3%, Test_loss:0.280 Epoch:46, Train_acc:99.8%, Train_loss:0.027, Test_acc:89.7%, Test_loss:0.271 Epoch:47, Train_acc:99.9%, Train_loss:0.024, Test_acc:89.0%, Test_loss:0.295 Epoch:48, Train_acc:100.0%, Train_loss:0.024, Test_acc:89.7%, Test_loss:0.279 Epoch:49, Train_acc:99.7%, Train_loss:0.026, Test_acc:88.8%, Test_loss:0.294 Epoch:50, Train_acc:99.9%, Train_loss:0.023, Test_acc:89.5%, Test_loss:0.323 Done
5.保存训练效果最好的模型参数
验证集的作用就是监督训练是否过拟合；一般默认验证集的损失值经历由下降到上升的阶段；
保存在验证集上损失最小的那个迭代模型，其泛化能力应该最好；