1. 对模型训练神经网络的理解:
(内容为b站“我是土堆”课程内容,观点仅代表个人理解,如果觉得有说的不对或者缺漏,麻烦在评论区批评指正,互相学习!)
神经网络可以当成是个黑盒子(函数),里面有很多未知数(待训练的参数),通过输入端输入数据(图像便是以数组形式输入),经过随机计算得出预测结果,由预测结果与设定真实结果进行loss损失值计算(计算误差),由梯度返回对随机设定的参数进行调整,这也就是训练过程,直到误差符合训练标准。
2. 模型训练神经网络训练流程:
- 先通过dataset读取图片(或)其他open读取方法
- Dataloder将所读取图片进行封装batch处理
- 建立模型,选择loss计算方法及优化器
- 图片经过模型计算得出与分类target各分数
- 通过loss计算得出误差,并通过loss.backward得出梯度,便于优化模型中各层权重、偏置参数
- 优化器通过梯度进行梯度下降法(优化的方法不止这个)进行参数优化,可通过debug看出每轮训练各层weight、bias数据进行更新
- 随着多轮训练后,loss也会随着变小,达到最佳
- 建立验证数据集进行验证本次训练准确率如何,验证是否能够使用
1.数据集来源:Cifar-10 https://www.cs.toronto.edu/~kriz/cifar.html
2.建立网络模型
class cnnmodel(nn.Module):
def __init__(self):
super(cnnmodel, self).__init__()
self.model = nn.Sequential(
nn.Conv2d(3, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 64, 5, 1, 2),
nn.MaxPool2d(2),
nn.Flatten(),
nn.Linear(64*4*4, 64),
nn.Linear(64, 10)
)
def forward(self, x):
x = self.model(x)
return x
- 提取最大特征
3channel,32*32 image经过5*5卷积核进行卷积生成32channel图片,在经过最大池化提取特征,依次反复3次,最后形成64channel,4*4 image
- 计算分类相似值
经过flat层与线性层算出图片与10target的拟合分数,分数高的就是对应的target
3.选择损失函数
loss_fn = nn.CrossEntropyLoss()
4.选择SGD优化器
learning_rate = 1e-2
optimizer = torch.optim.SGD(tudui.parameters(), lr=learning_rate)
5.开始训练
# 记录训练的次数
total_train_step = 0
# 记录测试的次数
total_test_step = 0
# 训练的轮数
epoch = 30
for i in range(epoch):
print("-------第 {} 轮训练开始-------".format(i+1))
# 训练步骤开始
cmodel.train()
for data in train_dataloader:
imgs, targets = data
if torch.cuda.is_available():
imgs = imgs.cuda()
targets = targets.cuda()
outputs = cmodel(imgs)
loss = loss_fn(outputs, targets)
# 优化器优化模型
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_train_step = 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
cmodel.eval()
total_test_loss = 0
total_accuracy = 0
with torch.no_grad():
for data in test_dataloader:
imgs, targets = data
if torch.cuda.is_available():
imgs = imgs.cuda()
targets = targets.cuda()
outputs = cmodel(imgs)
loss = loss_fn(outputs, targets)
total_test_loss = total_test_loss + loss.item()
accuracy = (outputs.argmax(1) == targets).sum()
total_accuracy = 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/test_data_size))
total_test_step = total_test_step + 1
if(i==29):
torch.save(cmodel, "cmodel_gpu_{}.pth".format(i))
print("模型已保存")
完整源码:
import torch
import torchvision
from PIL import Image
from torch import nn
# 准备数据集
from torch import nn
from torch.utils.data import DataLoader
train_data = torchvision.datasets.CIFAR10(root="D:/VScode/pyproject/nueralnetwork/pytorch-tutorial-master/data", train=True, transform=torchvision.transforms.ToTensor(),
download=True)
test_data = torchvision.datasets.CIFAR10(root="D:/VScode/pyproject/nueralnetwork/pytorch-tutorial-master/data", train=False, transform=torchvision.transforms.ToTensor(),
download=True)
# length 长度
train_data_size = len(train_data)
test_data_size = len(test_data)
print("训练数据集的长度为:{}".format(train_data_size))
print("测试数据集的长度为:{}".format(test_data_size))
# 利用 DataLoader 来加载数据集
train_dataloader = DataLoader(train_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)
class cnnmodel(nn.Module):
def __init__(self):
super(cnnmodel, self).__init__()
self.model = nn.Sequential(
nn.Conv2d(3, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 64, 5, 1, 2),
nn.MaxPool2d(2),
nn.Flatten(),
nn.Linear(64*4*4, 64),
nn.Linear(64, 10)
)
def forward(self, x):
x = self.model(x)
return x
cmodel = cnnmodel()
if torch.cuda.is_available():
cmodel = cmodel.cuda()
# 损失函数
loss_fn = nn.CrossEntropyLoss()
if torch.cuda.is_available():
loss_fn = loss_fn.cuda()
# 优化器
# learning_rate = 0.01
learning_rate = 1e-2
optimizer = torch.optim.SGD(cmodel.parameters(), lr=learning_rate)
# 记录训练的次数
total_train_step = 0
# 记录测试的次数
total_test_step = 0
# 训练的轮数
epoch = 30
for i in range(epoch):
print("-------第 {} 轮训练开始-------".format(i+1))
# 训练步骤开始
cmodel.train()
for data in train_dataloader:
imgs, targets = data
if torch.cuda.is_available():
imgs = imgs.cuda()
targets = targets.cuda()
outputs = cmodel(imgs)
loss = loss_fn(outputs, targets)
# 优化器优化模型
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_train_step = 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
cmodel.eval()
total_test_loss = 0
total_accuracy = 0
with torch.no_grad():
for data in test_dataloader:
imgs, targets = data
if torch.cuda.is_available():
imgs = imgs.cuda()
targets = targets.cuda()
outputs = cmodel(imgs)
loss = loss_fn(outputs, targets)
total_test_loss = total_test_loss + loss.item()
accuracy = (outputs.argmax(1) == targets).sum()
total_accuracy = 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/test_data_size))
total_test_step = total_test_step + 1
if(i==29):
torch.save(cmodel, "cmodel_gpu_{}.pth".format(i))
print("模型已保存")