pytorch学习笔记-Loss的使用、在神经网络中加入Loss、优化器(optimizer)的使用

博主最近真要累鼠了...

anyway上号更新一点，预计下周就能把该系列更学完了= =

Loss的使用

具体关注一下官网上对于形状的说明，如果报错就看看是不是形状不符合要求，其他没啥

python 复制代码

import torch
from torch.nn import L1Loss,MSELoss, CrossEntropyLoss
from torch import nn

#L1Loss
inputs = torch.tensor([1,2,3],dtype=torch.float32)
targets = torch.tensor([1,2,5],dtype=torch.float32)

l1_loss = L1Loss()
res = l1_loss(inputs, targets)

print("L1Loss:",res)

#MSE
mse_loss = MSELoss()
res = mse_loss(inputs, targets)

print("mseLoss:",res)

#crossentropy
x = torch.tensor([0.1,0.2,0.3])#每一类的概率
y = torch.tensor([1])#目标类别编号

# Input: Shape(N,C), N:bt_size C:class
x = torch.reshape(x,(1,3))
# print(x)

cross_loss = CrossEntropyLoss()
res = cross_loss(x,y)
print("crossentropyLoss:",res)

# L1Loss: tensor(0.6667)
# mseLoss: tensor(1.3333)
# crossentropyLoss: tensor(1.1019)

在神经网络中加入Loss

了解Loss的本质就是计算真值和目标值之间的差距后，怎么在神经网络中引入loss也蛮自然的：

python 复制代码

import torch
import torch.nn as nn
from torch.nn import Conv2d, MaxPool2d, Flatten, Linear, Sequential, CrossEntropyLoss
from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms, datasets
from torch.utils.data import DataLoader

data_transforms = transforms.Compose([
    transforms.ToTensor()
])

test_data = datasets.CIFAR10(root="./dataset",
                             train=False,
                             transform=data_transforms)

dataloader = DataLoader(test_data,
                        batch_size=64)

class myModule(nn.Module):
    def __init__(self):
        super().__init__()
        
        self.model1 = Sequential(
            Conv2d(3,32,5,padding=2),
            MaxPool2d(2),
            Conv2d(32,32,5,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

my_module = myModule()

loss = CrossEntropyLoss()

for data in dataloader:
    imgs, targets = data
    outputs = my_module(imgs)
    res_loss = loss(outputs,targets)
    print(res_loss)

优化器的使用

引入Loss的目的是为了更好的进行参数更新，因此需要引入优化器

事实上引入这一步后也就基本知道了模型如何进行训练了

一般在一次学习中就进行了多次更新，需要进行多次学习，注意的点就是每次梯度计算优化前需要先将上一轮计算得到的梯度清零，因为上一批次的对本次的结果意义不大，剩下的就是用法：

python 复制代码

import torch
import torch.nn as nn
from torch.nn import Conv2d, MaxPool2d, Flatten, Linear, Sequential, CrossEntropyLoss
from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms, datasets
from torch.utils.data import DataLoader
from torch import optim


data_transforms = transforms.Compose([
    transforms.ToTensor()
])

test_data = datasets.CIFAR10(root="./dataset",
                             train=False,
                             transform=data_transforms)

dataloader = DataLoader(test_data,
                        batch_size=64)

class myModule(nn.Module):
    def __init__(self):
        super().__init__()
        
        self.model1 = Sequential(
            Conv2d(3,32,5,padding=2),
            MaxPool2d(2),
            Conv2d(32,32,5,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

my_module = myModule()

loss = CrossEntropyLoss()

#设置优化器
optimizer = torch.optim.SGD(my_module.parameters(), lr=0.01)

for epoch in range(5):
    running_loss = 0.0 #计算每一个epoch的loss
    for data in dataloader:
        imgs, targets = data
        outputs = my_module(imgs)
        res_loss = loss(outputs,targets)
        # print(res_loss)
        running_loss+=res_loss
        optimizer.zero_grad()#每一次要重新清零上一轮的梯度
        res_loss.backward()
        optimizer.step()#进行优化loss
    print(running_loss)

# tensor(360.6527, grad_fn=<AddBackward0>)
# tensor(355.2374, grad_fn=<AddBackward0>)
# tensor(336.8181, grad_fn=<AddBackward0>)
# tensor(320.9179, grad_fn=<AddBackward0>)
# tensor(312.2012, grad_fn=<AddBackward0>)