1. 损失函数与反向传播
1.1 损失函数（loss function）
损失函数：判断模型训练的好坏标准，你训练了模型，总得有个指标来说明它好不好吧.jpg
import torch
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])
# 平均差，总和差
loss = torch.nn.L1Loss(reduction='sum')
result = loss(inputs, targets)
print(result)
# 平方差
loss_mse = torch.nn.MSELoss()
result = loss_mse(inputs, targets)
print(result)
# 交叉熵
x = torch.tensor([0.1, 0.2, 0.3])
y = torch.tensor([1])
x = torch.reshape(x, [1, 3])
loss_cross = torch.nn.CrossEntropyLoss()
result = loss_cross(x, y)
print(result)
1.2 反向传播(backward)
反向传播：根据损失函数，反向传播计算梯度，然后更新参数，使得损失函数最小化
import torch
import torchvision
from torchvision import transforms
from torch.utils.data import DataLoader
from torch import nn
from torch.utils.tensorboard import SummaryWriter
diy_transform = transforms.Compose([transforms.ToTensor()])
dataset = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=False, transform=diy_transform, download=False)
dataloader = DataLoader(dataset=dataset, batch_size=1, shuffle=False, num_workers=0, drop_last=False)
writer = SummaryWriter(log_dir="logs")
class Wsxk(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = torch.nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2, dilation=1)
self.maxpool1 = torch.nn.MaxPool2d(kernel_size=2)
self.conv2 = torch.nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2, dilation=1)
self.maxpool2 = torch.nn.MaxPool2d(kernel_size=2)
self.conv3 = torch.nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2, dilation=1)
self.maxpool3 = torch.nn.MaxPool2d(kernel_size=2)
self.flattern = torch.nn.Flatten()
self.linear1 = torch.nn.Linear(in_features=1024, out_features=64)
self.linear2 = torch.nn.Linear(in_features=64, out_features=10)
self.model1 = torch.nn.Sequential(torch.nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2, dilation=1),
torch.nn.MaxPool2d(kernel_size=2),
torch.nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2, dilation=1),
torch.nn.MaxPool2d(kernel_size=2),
torch.nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2, dilation=1),
torch.nn.MaxPool2d(kernel_size=2),
torch.nn.Flatten(),
torch.nn.Linear(in_features=1024, out_features=64),
torch.nn.Linear(in_features=64, out_features=10))
def forward(self, x):
# x = self.conv1(x)
# x = self.maxpool1(x)
# x = self.conv2(x)
# x = self.maxpool2(x)
# x = self.conv3(x)
# x = self.maxpool3(x)
# x = self.flattern(x)
# x = self.linear1(x)
# x = self.linear2(x)
x = self.model1(x)
return x
wsxk = Wsxk()
loss = nn.CrossEntropyLoss()
for data in dataloader:
imgs, targets = data
outputs = wsxk(imgs)
print(outputs)
print(targets)
result_loss = loss(outputs, targets)
print(result_loss)
result_loss.backward() #反向传播
2. 优化器torch.optim
优化器：根据梯度更新参数，使得损失函数最小化
import torch
import torchvision
from torchvision import transforms
from torch.utils.data import DataLoader
from torch import nn
from torch.utils.tensorboard import SummaryWriter
diy_transform = transforms.Compose([transforms.ToTensor()])
dataset = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=False, transform=diy_transform, download=False)
dataloader = DataLoader(dataset=dataset, batch_size=1, shuffle=False, num_workers=0, drop_last=False)
writer = SummaryWriter(log_dir="logs")
class Wsxk(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = torch.nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2, dilation=1)
self.maxpool1 = torch.nn.MaxPool2d(kernel_size=2)
self.conv2 = torch.nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2, dilation=1)
self.maxpool2 = torch.nn.MaxPool2d(kernel_size=2)
self.conv3 = torch.nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2, dilation=1)
self.maxpool3 = torch.nn.MaxPool2d(kernel_size=2)
self.flattern = torch.nn.Flatten()
self.linear1 = torch.nn.Linear(in_features=1024, out_features=64)
self.linear2 = torch.nn.Linear(in_features=64, out_features=10)
self.model1 = torch.nn.Sequential(torch.nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2, dilation=1),
torch.nn.MaxPool2d(kernel_size=2),
torch.nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2, dilation=1),
torch.nn.MaxPool2d(kernel_size=2),
torch.nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2, dilation=1),
torch.nn.MaxPool2d(kernel_size=2),
torch.nn.Flatten(),
torch.nn.Linear(in_features=1024, out_features=64),
torch.nn.Linear(in_features=64, out_features=10))
def forward(self, x):
# x = self.conv1(x)
# x = self.maxpool1(x)
# x = self.conv2(x)
# x = self.maxpool2(x)
# x = self.conv3(x)
# x = self.maxpool3(x)
# x = self.flattern(x)
# x = self.linear1(x)
# x = self.linear2(x)
x = self.model1(x)
return x
wsxk = Wsxk()
loss = nn.CrossEntropyLoss()
optim = torch.optim.SGD(wsxk.parameters(), lr=0.01)
# 训练模型 20轮
for epoch in range(20):
running_loss = 0.0
# 每一轮
for data in dataloader:
imgs, targets = data
outputs = wsxk(imgs)
result_loss = loss(outputs, targets)
optim.zero_grad()
result_loss.backward()
optim.step()
# print(result_loss)
running_loss += result_loss
print(running_loss)
3. 现有模型的使用以及修改
读作现有模型的使用与修改，写作炼丹
现有模型一般都集成在了库里，例如torchvision就集成了计算机视觉的一些现有模型，可供使用
import torch
from torch import nn
import torchvision
from torchvision import transforms
# 更改预训练模型的下载文件目录
import os
os.environ["TORCH_HOME"] = "C:/Users/11029/Documents/PythonProject/torchvision_dataset"
transform_diy = transforms.Compose([transforms.ToTensor()])
train_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=True, download=False)
# 没有经过预训练的模型
vgg16_false = torchvision.models.vgg16()
# 经过预训练的模型
vgg16_true = torchvision.models.vgg16(weights=torchvision.models.VGG16_Weights.IMAGENET1K_V1)
print(vgg16_true)
# 在现有模型下添加一层，所谓炼丹
vgg16_true.add_module("add_linear", module=nn.Linear(in_features=1000, out_features=10))
# vgg16_true.classifier.add_module("add_linear", module=nn.Linear(in_features=1000, out_features=10))
print(vgg16_true)
# 在现有模型的层数上做替换
vgg16_false.classifier[6] = nn.Linear(in_features=1000, out_features=10)
print(vgg16_false)
4. 现有模型的保存以及加载
import torch
import torchvision
vgg16 = torchvision.models.vgg16()
# 保存方式1 模型结构+参数
torch.save(vgg16, "vgg16_method1.pth")
# 保存方式2,只保持参数，不保存结构
torch.save(vgg16.state_dict(), "vgg16_method2.pth")
注意，在使用方式二保存模型时，必须确保加载时模型的定义（即类的定义能被访问到），否则会报错
import torch
import torchvision
# 使用方式1 模型结构+参数
model = torch.load("vgg16_method1.pth")
# print(model)
# 使用方式2 参数（官方推荐）
model2 = torch.load("vgg16_method2.pth")
vgg16 = torchvision.models.vgg16()
vgg16.load_state_dict(model2)
print(vgg16)
5. 模型的训练与测试
5.1 version 1
import torch
import torchvision
from torch.utils.data import DataLoader
from torch import nn
# 准备数据集
transorm_diy = torchvision.transforms.ToTensor()
train_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=True, transform=transorm_diy, download=True)
test_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=False, transform=transorm_diy, download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
print("训练集的长度为 {}".format(train_data_size))
print("测试集的长度为 {}".format(test_data_size))
# 加载数据集
train_dataloader = DataLoader(dataset=train_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
test_dataloader = DataLoader(dataset=test_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
# 搭建神经网络
class Wsxk(nn.Module):
def __init__(self):
super().__init__()
self.model = nn.Sequential(
nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Flatten(),
nn.Linear(in_features=64*4*4, out_features=64),
nn.Linear(in_features=64, out_features=10)
)
def forward(self, x):
x = self.model(x)
return x
# 创建模型
wsxk = Wsxk()
# 损失函数
loss_fn = nn.CrossEntropyLoss()
# 优化器 随机梯度下降
learning_rate = 1e-2
optimizer = torch.optim.SGD(params=wsxk.parameters(), lr=learning_rate)
# 设置训练网络的一些参数
# 记录训练次数
total_train_step = 0
# 记录测试次数
total_test_step = 0
# 记录训练论述
epoch = 10
for i in range(epoch):
print("----------第 {} 轮训练开始----------".format(i+1))
# 训练步骤开始
for data in train_dataloader:
imgs, targets = data
outputs = wsxk(imgs)
loss = loss_fn(outputs, targets)
# 优化器优化模型
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_train_step += 1
print("训练次数: {}, loss: {}".format(total_train_step, loss.item()))
# if __name__ == "__main__":
# wsxk = Wsxk()
# input = torch.ones(64, 3, 32, 32)
# output = wsxk(input)
# print(output.shape)
5.2 version 2
在version 1的基础上进行测试，得到version 2
import torch
import torchvision
from torch.utils.data import DataLoader
from torch import nn
from torch.utils.tensorboard import SummaryWriter
# 准备数据集
transorm_diy = torchvision.transforms.ToTensor()
train_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=True, transform=transorm_diy, download=True)
test_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=False, transform=transorm_diy, download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
print("训练集的长度为 {}".format(train_data_size))
print("测试集的长度为 {}".format(test_data_size))
# 加载数据集
train_dataloader = DataLoader(dataset=train_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
test_dataloader = DataLoader(dataset=test_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
# 搭建神经网络
class Wsxk(nn.Module):
def __init__(self):
super().__init__()
self.model = nn.Sequential(
nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Flatten(),
nn.Linear(in_features=64*4*4, out_features=64),
nn.Linear(in_features=64, out_features=10)
)
def forward(self, x):
x = self.model(x)
return x
# 创建模型
wsxk = Wsxk()
# 损失函数
loss_fn = nn.CrossEntropyLoss()
# 优化器 随机梯度下降
learning_rate = 1e-2
optimizer = torch.optim.SGD(params=wsxk.parameters(), lr=learning_rate)
# 设置训练网络的一些参数
# 记录训练次数
total_train_step = 0
# 记录测试次数
total_test_step = 0
# 记录训练论述
epoch = 10
# 画图
writer = SummaryWriter(log_dir="logs")
for i in range(epoch):
print("----------第 {} 轮训练开始----------".format(i+1))
# 训练步骤开始
for data in train_dataloader:
imgs, targets = data
outputs = wsxk(imgs)
loss = loss_fn(outputs, targets)
# 优化器优化模型
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_train_step += 1
if total_train_step % 100 == 0:
print("训练次数: {}, loss: {}".format(total_train_step, loss.item()))
writer.add_scalar("train_loss", loss.item(), total_train_step) # 两条线，深色曲线：平滑处理 浅色曲线：真实曲线
# 测试步骤开始
total_loss = 0
total_accuracy = 0
with torch.no_grad():
for data in test_dataloader:
imgs, targets = data
outputs = wsxk(imgs)
loss = loss_fn(outputs, targets)
total_loss += loss.item()
accuracy = (outputs.argmax(1) == targets).sum()
total_accuracy += accuracy
total_test_step += 1
print("第{}次整体测试集合上的loss：{}".format(total_test_step, total_loss))
print("第{}次整体测试集合的正确率：{}".format(total_test_step, total_accuracy/test_data_size))
writer.add_scalar("total_test_loss", total_loss, total_test_step)
writer.add_scalar("total_accuracy_rate", total_accuracy/test_data_size,total_test_step)
# 保存训练一次的模型
torch.save(wsxk, "model_save/wsxk_{}.pth".format(i))
writer.close()
# if __name__ == "__main__":
# wsxk = Wsxk()
# input = torch.ones(64, 3, 32, 32)
# output = wsxk(input)
# print(output.shape)
5.3 version 3
有些细节需要注意，有的代码会在测试步骤开始前调用eval()，在训练步骤开始前会调用train(),这两个函数的作用还是相对较少的，一般只对Dropout层和BatchNorm层有作用
import torch
import torchvision
from torch.utils.data import DataLoader
from torch import nn
from torch.utils.tensorboard import SummaryWriter
# 准备数据集
transorm_diy = torchvision.transforms.ToTensor()
train_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=True, transform=transorm_diy, download=True)
test_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=False, transform=transorm_diy, download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
print("训练集的长度为 {}".format(train_data_size))
print("测试集的长度为 {}".format(test_data_size))
# 加载数据集
train_dataloader = DataLoader(dataset=train_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
test_dataloader = DataLoader(dataset=test_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
# 搭建神经网络
class Wsxk(nn.Module):
def __init__(self):
super().__init__()
self.model = nn.Sequential(
nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Flatten(),
nn.Linear(in_features=64*4*4, out_features=64),
nn.Linear(in_features=64, out_features=10)
)
def forward(self, x):
x = self.model(x)
return x
# 创建模型
wsxk = Wsxk()
# 损失函数
loss_fn = nn.CrossEntropyLoss()
# 优化器 随机梯度下降
learning_rate = 1e-2
optimizer = torch.optim.SGD(params=wsxk.parameters(), lr=learning_rate)
# 设置训练网络的一些参数
# 记录训练次数
total_train_step = 0
# 记录测试次数
total_test_step = 0
# 记录训练论述
epoch = 10
# 画图
writer = SummaryWriter(log_dir="logs")
for i in range(epoch):
print("----------第 {} 轮训练开始----------".format(i+1))
# 训练步骤开始
wsxk.train()
for data in train_dataloader:
imgs, targets = data
outputs = wsxk(imgs)
loss = loss_fn(outputs, targets)
# 优化器优化模型
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_train_step += 1
if total_train_step % 100 == 0:
print("训练次数: {}, loss: {}".format(total_train_step, loss.item()))
writer.add_scalar("train_loss", loss.item(), total_train_step) # 两条线，深色曲线：平滑处理 浅色曲线：真实曲线
# 测试步骤开始
wsxk.eval()
total_loss = 0
total_accuracy = 0
with torch.no_grad():
for data in test_dataloader:
imgs, targets = data
outputs = wsxk(imgs)
loss = loss_fn(outputs, targets)
total_loss += loss.item()
accuracy = (outputs.argmax(1) == targets).sum()
total_accuracy += accuracy
total_test_step += 1
print("第{}次整体测试集合上的loss：{}".format(total_test_step, total_loss))
print("第{}次整体测试集合的正确率：{}".format(total_test_step, total_accuracy/test_data_size))
writer.add_scalar("total_test_loss", total_loss, total_test_step)
writer.add_scalar("total_accuracy_rate", total_accuracy/test_data_size,total_test_step)
# 保存训练一次的模型
torch.save(wsxk, "model_save/wsxk_{}.pth".format(i))
writer.close()
# if __name__ == "__main__":
# wsxk = Wsxk()
# input = torch.ones(64, 3, 32, 32)
# output = wsxk(input)
# print(output.shape)
5.4 version 4:利用GPU加速训练: 方式一
其实刚刚的训练代码中，没有使用cuda进行加速
使用cuda进行加速的方法其实很简单，方式一：只要对模型，模型的输入和标注，损失函数进行.cuda()即可。
使用GPU进行运算，你会发现运算快了很多,肉眼可见的加速
import torch
import torchvision
from torch.utils.data import DataLoader
from torch import nn
from torch.utils.tensorboard import SummaryWriter
# 准备数据集
transorm_diy = torchvision.transforms.ToTensor()
train_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=True, transform=transorm_diy, download=True)
test_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=False, transform=transorm_diy, download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
print("训练集的长度为 {}".format(train_data_size))
print("测试集的长度为 {}".format(test_data_size))
# 加载数据集
train_dataloader = DataLoader(dataset=train_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
test_dataloader = DataLoader(dataset=test_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
# 搭建神经网络
class Wsxk(nn.Module):
def __init__(self):
super().__init__()
self.model = nn.Sequential(
nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Flatten(),
nn.Linear(in_features=64*4*4, out_features=64),
nn.Linear(in_features=64, out_features=10)
)
def forward(self, x):
x = self.model(x)
return x
# 创建模型
wsxk = Wsxk()
if torch.cuda.is_available():
wsxk = wsxk.cuda() # cuda 利用GPU加速模型运算
# 损失函数
loss_fn = nn.CrossEntropyLoss()
if torch.cuda.is_available():
loss_fn = loss_fn.cuda() # cuda 利用GPU加速损失函数计算
# 优化器 随机梯度下降
learning_rate = 1e-2
optimizer = torch.optim.SGD(params=wsxk.parameters(), lr=learning_rate)
# 设置训练网络的一些参数
# 记录训练次数
total_train_step = 0
# 记录测试次数
total_test_step = 0
# 记录训练论述
epoch = 10
# 画图
writer = SummaryWriter(log_dir="logs")
for i in range(epoch):
print("----------第 {} 轮训练开始----------".format(i+1))
# 训练步骤开始
wsxk.train()
for data in train_dataloader:
imgs, targets = data
# 利用cuda加速imgs，targets 喂入模型
if torch.cuda.is_available():
imgs = imgs.cuda()
targets = targets.cuda()
outputs = wsxk(imgs)
loss = loss_fn(outputs, targets)
# 优化器优化模型
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_train_step += 1
if total_train_step % 100 == 0:
print("训练次数: {}, loss: {}".format(total_train_step, loss.item()))
writer.add_scalar("train_loss", loss.item(), total_train_step) # 两条线，深色曲线：平滑处理 浅色曲线：真实曲线
# 测试步骤开始
wsxk.eval()
total_loss = 0
total_accuracy = 0
with torch.no_grad():
for data in test_dataloader:
imgs, targets = data
# 利用cuda加速imgs， targets 喂入模型
if torch.cuda.is_available():
imgs = imgs.cuda()
targets = targets.cuda()
outputs = wsxk(imgs)
loss = loss_fn(outputs, targets)
total_loss += loss.item()
accuracy = (outputs.argmax(1) == targets).sum()
total_accuracy += accuracy
total_test_step += 1
print("第{}次整体测试集合上的loss：{}".format(total_test_step, total_loss))
print("第{}次整体测试集合的正确率：{}".format(total_test_step, total_accuracy/test_data_size))
writer.add_scalar("total_test_loss", total_loss, total_test_step)
writer.add_scalar("total_accuracy_rate", total_accuracy/test_data_size,total_test_step)
# 保存训练一次的模型
torch.save(wsxk, "model_save/wsxk_{}.pth".format(i))
writer.close()
# if __name__ == "__main__":
# wsxk = Wsxk()
# input = torch.ones(64, 3, 32, 32)
# output = wsxk(input)
# print(output.shape)
5.5 version 5:利用GPU加速训练: 方式二
第二种办法是定义一个device，然后将模型，模型的输入和标注，损失函数，都放到device上，这样就可以利用GPU加速了
import torch
import torchvision
from torch.utils.data import DataLoader
from torch import nn
from torch.utils.tensorboard import SummaryWriter
# 定义训练的设备
device = torch.device("cuda:0")
# 准备数据集
transorm_diy = torchvision.transforms.ToTensor()
train_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=True, transform=transorm_diy, download=True)
test_data = torchvision.datasets.CIFAR10(root="torchvision_dataset", train=False, transform=transorm_diy, download=True)
train_data_size = len(train_data)
test_data_size = len(test_data)
print("训练集的长度为 {}".format(train_data_size))
print("测试集的长度为 {}".format(test_data_size))
# 加载数据集
train_dataloader = DataLoader(dataset=train_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
test_dataloader = DataLoader(dataset=test_data, batch_size=64, shuffle=False, num_workers=0, drop_last=False)
# 搭建神经网络
class Wsxk(nn.Module):
def __init__(self):
super().__init__()
self.model = nn.Sequential(
nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=2),
nn.Flatten(),
nn.Linear(in_features=64*4*4, out_features=64),
nn.Linear(in_features=64, out_features=10)
)
def forward(self, x):
x = self.model(x)
return x
# 创建模型
wsxk = Wsxk()
wsxk = wsxk.to(device) # cuda 利用GPU加速模型运算
# 损失函数
loss_fn = nn.CrossEntropyLoss()
loss_fn = loss_fn.to(device) # cuda 利用GPU加速损失函数计算
# 优化器 随机梯度下降
learning_rate = 1e-2
optimizer = torch.optim.SGD(params=wsxk.parameters(), lr=learning_rate)
# 设置训练网络的一些参数
# 记录训练次数
total_train_step = 0
# 记录测试次数
total_test_step = 0
# 记录训练论述
epoch = 10
# 画图
writer = SummaryWriter(log_dir="logs")
for i in range(epoch):
print("----------第 {} 轮训练开始----------".format(i+1))
# 训练步骤开始
wsxk.train()
for data in train_dataloader:
imgs, targets = data
# 利用cuda加速imgs，targets 喂入模型
imgs = imgs.to(device)
targets = targets.to(device)
outputs = wsxk(imgs)
loss = loss_fn(outputs, targets)
# 优化器优化模型
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_train_step += 1
if total_train_step % 100 == 0:
print("训练次数: {}, loss: {}".format(total_train_step, loss.item()))
writer.add_scalar("train_loss", loss.item(), total_train_step) # 两条线，深色曲线：平滑处理 浅色曲线：真实曲线
# 测试步骤开始
wsxk.eval()
total_loss = 0
total_accuracy = 0
with torch.no_grad():
for data in test_dataloader:
imgs, targets = data
# 利用cuda加速imgs， targets 喂入模型
imgs = imgs.to(device)
targets = targets.to(device)
outputs = wsxk(imgs)
loss = loss_fn(outputs, targets)
total_loss += loss.item()
accuracy = (outputs.argmax(1) == targets).sum()
total_accuracy += accuracy
total_test_step += 1
print("第{}次整体测试集合上的loss：{}".format(total_test_step, total_loss))
print("第{}次整体测试集合的正确率：{}".format(total_test_step, total_accuracy/test_data_size))
writer.add_scalar("total_test_loss", total_loss, total_test_step)
writer.add_scalar("total_accuracy_rate", total_accuracy/test_data_size,total_test_step)
# 保存训练一次的模型
torch.save(wsxk, "model_save/wsxk_{}.pth".format(i))
writer.close()
# if __name__ == "__main__":
# wsxk = Wsxk()
# input = torch.ones(64, 3, 32, 32)
# output = wsxk(input)
# print(output.shape)