Pytorch学习笔记(四)
一、损失函数与反向传播
1、L1Loss
结果为输出与标签之间的差的和平均值或是和
inputs = torch.tensor([1,2,3], dtype = torch.float32)
targets = torch.tensor([1,2,5],dtype = torch.float32)
inputs = torch.reshape(inputs, (1,1,1,3))
targets = torch.reshape(targets, (1,1,1,3))
loss = L1Loss(reduction='sum') #reduction决定是和的平均值还是和
result = loss(inputs,targets)
print(result)
输出为2 = 0 + 0 + 2
2、MSELoss
结果为输出与标签之间差平方的和平均值或是和
loss_mse = nn.MESLoss(reduction='sum')
result_mse = loss_mse(inputs, targets)
print(result_mse)
输出为4 = 0 + 0 + 2^2
3、CrossEntropyLoss
要求分类问题,并且输出为每个类别的概率,而标签为类别的序号
x = torch.tensor([0.1, 0.2, 0.3])
y = torch.tensor([1])
x = torch.reshape(x, (1, 3))
loss_cross = loss_cross(x, y)
print(result_cross)
输出为1.1019 = -0.2 + ln(exp(0.1) + exp(0.2) + exp(0.3))
4、将loss应用到神经网络中
dataset = torchvision.datasets.CIFAR10("./dataset", train=False, transform=torchvision.transform.ToTensor(), download=True)
dataloader = DataLoader(dataset,batch_size=64)
class Test(nn.Module):
def __init__(self):
super(Test, self).__init__()
#使用Sequential提高代码的简洁可读性
self.model1 = Sequential{
Conv2d(3, 32, 5, padding=2),
MaxPool2d(2),
Conv2d(32, 64, 5, padding=2),
MaxPool2d(2),
Conv2d(32, 64, 5, padding=2),
MaxPool2d(2),
Flatten(),
Linear(1024, 64),
Linear(64, 10)
)
def forward(self, x):
x = self.model1(x)
return x
loss = nn.CrossEntropyLoss()
test = Test()
for data in datloader:
imgs, targets = data
outputs = test(imgs)
result_loss = loss(outputs, targets)
result_loss.backward() #反向传播,此时查看节点的参数,grad中发现梯度数据
5、反向传播优化器
优化器根据梯度调整参数,达到降低误差的作用
主要参数:params和lr(学习速率)
dataset = torchvision.datasets.CIFAR10("./dataset", train=False, transform=torchvision.transform.ToTensor(), download=True)
dataloader = DataLoader(dataset,batch_size=64)
class Test(nn.Module):
def __init__(self):
super(Test, self).__init__()
#使用Sequential提高代码的简洁可读性
self.model1 = Sequential{
Conv2d(3, 32, 5, padding=2),
MaxPool2d(2),
Conv2d(32, 64, 5, padding=2),
MaxPool2d(2),
Conv2d(32, 64, 5, padding=2),
MaxPool2d(2),
Flatten(),
Linear(1024, 64),
Linear(64, 10)
)
def forward(self, x):
x = self.model1(x)
return x
loss = nn.CrossEntropyLoss()
test = Test()
optim = torch.optim.SGD(test.parameters(), lr=0.01) #创建优化器
for epoch in range(20): #迭代优化
for data in datloader:
imgs, targets = data
outputs = test(imgs)
result_loss = loss(outputs, targets)
optim.zero_grad() #优化器中的参数清零,必不可少!
result_loss.backward() #反向传播,此时查看节点的参数,grad中发现梯度数据
optim.step() #使用优化器更新参数
二、网络模型
1、现有网络模型的使用和修改
主要使用VGG模型来进行讲解
参数:
pretrained(bool):下载的网络模型参数是否在数据集中训练好,即为true时模型中的参数为训练好的,而false时参数为初始化的参数
progress(bool):是否显示下载记录条
vgg16 = torchvision.models.vgg16(pretrained=True)
dataset = torchvision.datasets.CIFAR10("./dataset", train=True, transform=torchvision.transform.ToTensor(), download=True)
vgg16.add_module('add_linear', nn.Linear(1000, 10)) #在VGG16最后添加一个线性层
vgg16.classifier[6] = nn.Linear(4096, 10) #修改vgg16中classifier中的第六层
2、网络模型的保存与读取
vgg16 = torchvision.models.vgg16(pretrained=False)
#保存方式1 不仅保存了模型,同时保存了模型的参数
torch.save(vgg16, "vgg16_method1.pth") #模型保存名称,后缀最好使用.pth
#保存方式2(官方推荐) 将vgg网络模型的参数保存成字典形式
torch.save(vgg16.state_dict(),"vgg16_method2.pth")
#加载模型-》保存方式1
model = torch.load("vgg16_method1.pth")
#加载模型-》保存方式2
vgg16 = torchvision.models.vgg16(pretrained=False)
vgg16.load_state_dict(torch.load("vgg16_method2.pth"))
完整的模型训练套路
#准备数据集
train_data = torchvision.datasets.CIFAR10(root="./dataset", train=True, transform=torchvision.transforms.ToTensor(), download=True)
test_data = torchvision.datasets.CIFAR10(root="./dataset", train=False, transform=torchvision.transforms.ToTensor(), download=True)
#加载数据集
train_dataloader = DataLoader(train_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)
#搭建神经网络,可以将这一部分的代码单独放到一个model的python中,model文件应该和本文件在一个文件夹下
class Test(nn.Module):
def __init__(self):
super(Test, self).__init__()
#使用Sequential提高代码的简洁可读性
self.model1 = Sequential{
Conv2d(3, 32, 5, padding=2),
MaxPool2d(2),
Conv2d(32, 64, 5, padding=2),
MaxPool2d(2),
Conv2d(32, 64, 5, padding=2),
MaxPool2d(2),
Flatten(),
Linear(1024, 64),
Linear(64, 10)
)
def forward(self, x):
x = self.model1(x)
return x
#创建网络模型
test = Test()
#损失函数
loss_fn = nn.CrossEntropyLoss()
#优化器
learning_rate = 0.01 #或为1e-2
optimizer = torch.optim.SGD(test.parameters(), lr = learning_rate)
#设置训练网络的参数
total_train_step = 0 #记录训练的次数
total_test_step = 0 #记录测试的次数
epoch = 10 #训练的迭代次数
for i in range(epoch):
print("---------第{}轮训练开始----------".format(i+1))
#训练步骤开始
test.train() #只对某些特定的层有作用
for data in train_dataloader:
imgs, targets = data
outputs = test(imgs)
loss = loss_fn(outputs, targets)
#优化器优化模型
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_train_setp = total_train_step + 1
if total_train_step % 100 == 0
print("训练次数:{}, loss:{}".format(total_train_step, loss.item()))
#测试步骤开始
test.eval() #只对某些特定的层有作用
total_test_loss = 0
total_accuracy = 0
with torch.no_grad():
for data in test_dataloader:
imgs, targets = data
outputs = test(imgs)
loss = loss_fn(outputs, targets)
total_test_loss = total_test_loss + loss.item()
accuracy = (outputs.argmax(1) == targets).sum() #argmax将输出的概率值转化成1/0
total_accuacy = total_accuracy + accuracy
print("整体测试集上的Loss:{}".format(total_test_loss))
print("整体测试集上的正确率:{}".format(total_accuracy/len(test_data)))
torch.save(test, "test_{}.pth".format(i))