- 🍨 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
🍺要求：
- 了解如何设置动态学习率（重点）
- 调整代码使测试集accuracy到达84%。
一、导入库
import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets
import torch.nn.functional as F
import os,PIL,pathlib
import matplotlib.pyplot as plt
import warnings
import os,PIL,random,pathlib
二、导入数据
data_dir = './data/shoot'
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("\\")[2] for path in data_paths]
#print(classeNames)
train_transforms = transforms.Compose([
transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸
# transforms.RandomHorizontalFlip(), # 随机水平翻转
transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])
test_transform = transforms.Compose([
transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸
transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])
train_dataset = datasets.ImageFolder("./data/shoot/train/",transform=train_transforms)
test_dataset = datasets.ImageFolder("./data/shoot/test/",transform=test_transform)
#print(train_dataset.class_to_idx) #查看图片标签
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=True,
num_workers=1)
三、构建CNN网络模型
class Model(nn.Module):
def __init__(self):
super(Model, self).__init__()
self.conv1 = nn.Sequential(
nn.Conv2d(3, 12, kernel_size=5, padding=0), # 12*220*220
nn.BatchNorm2d(12),
nn.ReLU())
self.conv2 = nn.Sequential(
nn.Conv2d(12, 12, kernel_size=5, padding=0), # 12*216*216
nn.BatchNorm2d(12),
nn.ReLU())
self.pool3 = nn.Sequential(
nn.MaxPool2d(2)) # 12*108*108
self.conv4 = nn.Sequential(
nn.Conv2d(12, 24, kernel_size=5, padding=0), # 24*104*104
nn.BatchNorm2d(24),
nn.ReLU())
self.conv5 = nn.Sequential(
nn.Conv2d(24, 24, kernel_size=5, padding=0), # 24*100*100
nn.BatchNorm2d(24),
nn.ReLU())
self.pool6 = nn.Sequential(
nn.MaxPool2d(2)) # 24*50*50
self.dropout = nn.Sequential(
nn.Dropout(0.2))
self.fc = nn.Sequential(
nn.Linear(24 * 50 * 50, len(classeNames)))
def forward(self, x):
batch_size = x.size(0)
x = self.conv1(x) # 卷积-BN-激活
x = self.conv2(x) # 卷积-BN-激活
x = self.pool3(x) # 池化
x = self.conv4(x) # 卷积-BN-激活
x = self.conv5(x) # 卷积-BN-激活
x = self.pool6(x) # 池化
x = self.dropout(x)
x = x.view(batch_size, -1) # flatten 变成全连接网络需要的输入 (batch, 24*50*50) ==> (batch, -1), -1 此处自动算出的是24*50*50
x = self.fc(x)
return x
四、训练函数
# 训练循环
def train(dataloader, model, loss_fn, optimizer):
size = len(dataloader.dataset) # 训练集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)
train_loss, train_acc = 0, 0 # 初始化训练损失和正确率
for X, y in dataloader: # 获取图片及其标签
X, y = X.to(device), y.to(device)
# 计算预测误差
pred = model(X) # 网络输出
loss = loss_fn(pred, y) # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失
# 反向传播
optimizer.zero_grad() # grad属性归零
loss.backward() # 反向传播
optimizer.step() # 每一步自动更新
# 记录acc与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
五、测试函数
def test (dataloader, model, loss_fn):
size = len(dataloader.dataset) # 测试集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)
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_loss
六、设置学习率
def adjust_learning_rate(optimizer, epoch, start_lr):
# 每 2 个epoch衰减到原来的 0.92
lr = start_lr * (0.92 ** (epoch // 2))
for param_group in optimizer.param_groups:
param_group['lr'] = lr
learn_rate = 1e-4 # 初始学习率
optimizer = torch.optim.SGD(model.parameters(), lr=learn_rate)
七、主函数以及可视化展示
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
epochs = 30
train_loss = []
train_acc = []
test_loss = []
test_acc = []
if __name__ == '__main__':
for epoch in range(epochs):
# 更新学习率（使用自定义学习率时使用）
adjust_learning_rate(optimizer, epoch, learn_rate)
model.train()
epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer)
# scheduler.step() # 更新学习率（调用官方动态学习率接口时使用）
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)
# 获取当前的学习率
lr = optimizer.state_dict()['param_groups'][0]['lr']
template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}')
print(template.format(epoch + 1, epoch_train_acc * 100, epoch_train_loss,
epoch_test_acc * 100, epoch_test_loss, lr))
print('Done')
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
epochs_range = range(epochs)
plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, train_acc, label='Training Accuracy')
plt.plot(epochs_range, test_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
plt.subplot(1, 2, 2)
plt.plot(epochs_range, train_loss, label='Training Loss')
plt.plot(epochs_range, test_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
八、调用模型
classes = list(total_data.class_to_idx)
predict_one_image(image_path='./data/1.jpg',
model=model,
transform=train_transforms,
classes=classes)
结果
Epoch: 1, Train_acc:59.4%, Train_loss:0.781, Test_acc:52.6%, Test_loss:0.651, Lr:1.00E-04
Epoch: 2, Train_acc:79.3%, Train_loss:0.457, Test_acc:75.0%, Test_loss:0.560, Lr:1.00E-04
Epoch: 3, Train_acc:86.5%, Train_loss:0.309, Test_acc:78.9%, Test_loss:0.504, Lr:1.00E-04
Epoch: 4, Train_acc:92.0%, Train_loss:0.257, Test_acc:76.3%, Test_loss:0.528, Lr:9.20E-05
Epoch: 5, Train_acc:96.8%, Train_loss:0.166, Test_acc:80.3%, Test_loss:0.456, Lr:9.20E-05
Epoch: 6, Train_acc:97.8%, Train_loss:0.133, Test_acc:80.3%, Test_loss:0.470, Lr:9.20E-05
Epoch: 7, Train_acc:98.6%, Train_loss:0.104, Test_acc:81.6%, Test_loss:0.419, Lr:8.46E-05
Epoch: 8, Train_acc:99.4%, Train_loss:0.091, Test_acc:82.9%, Test_loss:0.418, Lr:8.46E-05
Epoch: 9, Train_acc:99.6%, Train_loss:0.069, Test_acc:81.6%, Test_loss:0.420, Lr:8.46E-05
Epoch:10, Train_acc:99.6%, Train_loss:0.060, Test_acc:81.6%, Test_loss:0.558, Lr:7.79E-05
Epoch:11, Train_acc:100.0%, Train_loss:0.049, Test_acc:81.6%, Test_loss:0.435, Lr:7.79E-05
Epoch:12, Train_acc:100.0%, Train_loss:0.042, Test_acc:77.6%, Test_loss:0.402, Lr:7.79E-05
Epoch:13, Train_acc:100.0%, Train_loss:0.040, Test_acc:82.9%, Test_loss:0.442, Lr:7.16E-05
Epoch:14, Train_acc:99.8%, Train_loss:0.038, Test_acc:80.3%, Test_loss:0.534, Lr:7.16E-05
Epoch:15, Train_acc:100.0%, Train_loss:0.036, Test_acc:81.6%, Test_loss:0.524, Lr:7.16E-05
Epoch:16, Train_acc:100.0%, Train_loss:0.027, Test_acc:80.3%, Test_loss:0.475, Lr:6.59E-05
Epoch:17, Train_acc:100.0%, Train_loss:0.028, Test_acc:81.6%, Test_loss:0.492, Lr:6.59E-05
Epoch:18, Train_acc:100.0%, Train_loss:0.026, Test_acc:80.3%, Test_loss:0.589, Lr:6.59E-05
Epoch:19, Train_acc:100.0%, Train_loss:0.023, Test_acc:81.6%, Test_loss:0.455, Lr:6.06E-05
Epoch:20, Train_acc:100.0%, Train_loss:0.021, Test_acc:81.6%, Test_loss:0.473, Lr:6.06E-05
Epoch:21, Train_acc:100.0%, Train_loss:0.021, Test_acc:84.2%, Test_loss:0.439, Lr:6.06E-05
Epoch:22, Train_acc:100.0%, Train_loss:0.019, Test_acc:80.3%, Test_loss:0.440, Lr:5.58E-05
Epoch:23, Train_acc:100.0%, Train_loss:0.017, Test_acc:81.6%, Test_loss:0.490, Lr:5.58E-05
Epoch:24, Train_acc:100.0%, Train_loss:0.015, Test_acc:84.2%, Test_loss:0.542, Lr:5.58E-05
Epoch:25, Train_acc:100.0%, Train_loss:0.016, Test_acc:82.9%, Test_loss:0.502, Lr:5.13E-05
Epoch:26, Train_acc:100.0%, Train_loss:0.015, Test_acc:84.2%, Test_loss:0.413, Lr:5.13E-05
Epoch:27, Train_acc:100.0%, Train_loss:0.015, Test_acc:84.2%, Test_loss:0.567, Lr:5.13E-05
Epoch:28, Train_acc:100.0%, Train_loss:0.013, Test_acc:82.9%, Test_loss:0.613, Lr:4.72E-05
Epoch:29, Train_acc:100.0%, Train_loss:0.012, Test_acc:82.9%, Test_loss:0.603, Lr:4.72E-05
Epoch:30, Train_acc:100.0%, Train_loss:0.012, Test_acc:84.2%, Test_loss:0.462, Lr:4.72E-05
Done
预测结果是：adidas
用Adam优化器
用SGD优化器
一开始用的SGD优化器但是准确率一直不好，所以改用了Adam优化器，结果提升很大。
本周也学习了动态学习率的调整，调整学习率的方法有以下几种：
固定学习率：在训练开始时，可以使用固定的学习率以训练，是最简单的一种。
学习率衰减：随着训练的进行，可以逐渐降低学习率，使网络收敛。学习率衰减可以通过以下几种方式实现：
固定衰减：在训练的每个epoch或一定的步数后，将学习率乘以一个衰减因子，例如0.1或0.5。
指数衰减：学习率按指数函数进行衰减，例如每个epoch或一定的步数后，将学习率乘以一个小于1的指数因子。
周期性衰减：学习率按周期性函数进行衰减，例如使用三角函数或余弦函数来调整学习率。
学习率自适应：可以使用自适应的学习率调整方法，根据模型的表现动态地调整学习率。常见的自适应学习率调整方法包括：
AdaGrad：根据参数的梯度历史信息来调整学习率，对于稀疏梯度的参数，学习率会相应地增大。
RMSProp：类似于AdaGrad，但对梯度历史信息进行指数加权平均，以减小历史梯度对学习率的影响。
Adam：结合了momentum和RMSProp的优点，使用梯度的一阶矩估计和二阶矩估计来调整学习率。
等间隔调整学习率（StepLR）：每隔一定周期（step_size）就对学习率按gamma参数进行一次衰减。
按需调整学习率（MultiStepLR）：按设定的间隔调整学习率，适合后期调试使用，观察loss曲线，为每个实验定制学习率调整时机。
指数衰减调整（ExponentialLR）：学习率按照指数函数进行衰减。
余弦退火（CosineAnnealingLR）：学习率按照余弦函数进行调整，常用于模拟退火算法中的学习率调整。
自适应调整学习率（ReduceLROnPlateau）：当模型的验证指标（如loss）不再改善时，降低学习率。
自定义调整学习率（LambdaLR）：根据自定义的函数动态调整学习率。
扫一扫
1262
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
