目录
程序代码
import torch
from torchvision import transforms
from torchvision import datasets
from torch.utils.data import DataLoader
import torch.nn.functional as F
import torch.optim as optim
batch_size = 64
transform = transforms.Compose([
transforms.ToTensor(), #将shape为(H, W, C)的img转为shape为(C, H, W)的tensor,将每一个数值归一化到[0,1]
transforms.Normalize((0.1307, ), (0.3081, )) #按通道进行数据标准化
])
train_dataset = datasets.MNIST(root = '../Pycharm/dataset/mnist/', train = True, download = True, transform = transform)
train_loader = DataLoader(train_dataset, shuffle = True, batch_size = batch_size)
test_dataset = datasets.MNIST(root = '../Pycharm/dataset/mnist/', train = False, download = True, transform = transform)
test_loader = DataLoader(test_dataset, shuffle = False, batch_size = batch_size)
class InceptionA(torch.nn.Module):
def __init__(self, in_channels):
super(InceptionA, self).__init__()
self.branch1x1 = torch.nn.Conv2d(in_channels, 16, kernel_size = 1) # 1x1卷积
self.branch5x5_1 = torch.nn.Conv2d(in_channels, 16, kernel_size = 1) # 先1x1卷积
self.branch5x5_2 = torch.nn.Conv2d(16, 24, kernel_size = 5, padding = 2) # 再5x5卷积 padding = 2是为了保证图像尺寸不变
self.branch3x3_1 = torch.nn.Conv2d(in_channels, 16, kernel_size = 1) # 先1x1卷积
self.branch3x3_2 = torch.nn.Conv2d(16, 24, kernel_size = 3, padding = 1) # 再3x3卷积
self.branch3x3_3 = torch.nn.Conv2d(24, 24, kernel_size = 3, padding = 1) # # 再3x3卷积 注意输入输出维度大小
self.branch_pool = torch.nn.Conv2d(in_channels, 24, kernel_size = 1) # 池化后再1x1卷积
def forward(self, x):
branch1x1 = self.branch1x1(x) # Module1
branch5x5 = self.branch5x5_1(x) # Module2
branch5x5 = self.branch5x5_2(branch5x5)
branch3x3 = self.branch3x3_1(x) # Module3
branch3x3 = self.branch3x3_2(branch3x3)
branch3x3 = self.branch3x3_3(branch3x3)
branch_pool = F.avg_pool2d(x, kernel_size = 3, stride = 1, padding = 1) # 平均池化
branch_pool = self.branch_pool(branch_pool) # Module4
outputs = [branch1x1, branch5x5, branch3x3, branch_pool]
return torch.cat(outputs, dim = 1)
class Net(torch.nn.Module):
def __init__(self): # 构造函数
super(Net, self).__init__()
self.conv1 = torch.nn.Conv2d(1, 10, kernel_size = 5) # 卷积层1
self.conv2 = torch.nn.Conv2d(88, 20, kernel_size = 5) # 卷积层2
self.incep1 = InceptionA(in_channels = 10)
self.incep2 = InceptionA(in_channels = 20)
self.mp = torch.nn.MaxPool2d(2)
self.fc = torch.nn.Linear(1408, 10)
def forward(self, x):
in_size = x.size(0)
x = F.relu(self.mp(self.conv1(x))) # 卷积、池化、激活函数
x = self.incep1(x)
x = F.relu(self.mp(self.conv2(x))) # 卷积、池化、激活函数
x = self.incep2(x)
x = x.view(in_size, -1) # reshape
x = self.fc(x) # 全连接层
return x
model = Net()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") # 调用GPU或CPU
model.to(device)
criterion = torch.nn.CrossEntropyLoss() # 计算交叉熵损失
optimizer = optim.SGD(model.parameters(), lr = 0.01, momentum = 0.5) #构建优化器,lr为学习率,momentum为冲量因子
def train(epoch):
running_loss = 0.0
for batch_idx, data in enumerate(train_loader, 0): # 遍历函数,0表示从第0个元素开始,返回数据下标和数值
inputs, target = data #特征,标签
inputs, target = inputs.to(device), target.to(device)
optimizer.zero_grad() #梯度归零
# forward + backward + updata
outputs = model(inputs)
loss = criterion(outputs, target) #计算损失
loss.backward() #反向传播梯度值
optimizer.step() #更新参数
running_loss += loss.item() #得到元素张量的一个元素值,将张量转换成浮点数
if batch_idx % 300 == 299:
print('[%d, %5d] loss: %.3f' % (epoch + 1, batch_idx + 1, running_loss / 300))
running_loss = 0.0
def test():
correct = 0
total = 0
with torch.no_grad(): #数据不计算梯度
for data in test_loader:
images, labels = data
images, labels = images.to(device), labels.to(device)
outputs = model(images)
_, predicted = torch.max(outputs.data, dim = 1) #predicted为tensor每行最大值的索引
total += labels.size(0) # 总样本
correct += (predicted == labels).sum().item() #预测准确的样本数
print('Accuracy on test set: %d %%' % (100 * correct / total)) #准确率
def main():
for epoch in range(10):
train(epoch)
test()
main()
Inception框图