基于pytorch的手写数字深度学习

#基于pytorch的数字识别

#作者：申炜 2023.03

#所用库————用ALT+ENTER引入

import torchvision.datasets

from torch.utils.data import DataLoader

from torch.utils.tensorboard import SummaryWriter

from model_number import *

#GPU运行

device=torch.device('cuda' if torch.cuda.is_available() else 'cpu')

#准备数据集MNIST

train_data=torchvision.datasets.MNIST(root='./data', train=True,transform=torchvision.transforms.ToTensor(),download=True)

test_data=torchvision.datasets.MNIST(root='./data', train=False,transform=torchvision.transforms.ToTensor(),download=True)

#数据集长度

train_data_size=len(train_data)

test_data_size=len(test_data)

print('训练数据集的长度为：{}，测试数据集的长度为：{}'.format(train_data_size,test_data_size))

#加载数据集

train_dataloader=DataLoader(train_data,batch_size=100,shuffle=True)

test_dataloader=DataLoader(test_data,batch_size=100,shuffle=True)

#深度学习网络

#图片大小为1*28*28，batch_size=100

import torch

from torch import nn

from torch.nn import Conv2d, MaxPool2d, Sequential, Flatten, Linear, ReLU, Dropout, Softmax

class Shen(nn.Module):

def __init__(self):

super(Shen,self).__init__()

self.model=Sequential(

Conv2d(1,16,kernel_size=3,padding=1),#100*16*28*28

ReLU(),

Conv2d(16,16,kernel_size=3,padding=1),#100*16*28*28

ReLU(),

MaxPool2d(2),#100*16*14*14

Conv2d(16,32,kernel_size=3,padding=1),#100*32*14*14

ReLU(),

Conv2d(32,32,kernel_size=3,padding=2),#100*32*16*16

ReLU(),

MaxPool2d(2),#100*32*8*8

Conv2d(32,64,kernel_size=3,padding=1),#100*64*8*8

ReLU(),

Conv2d(64,64,kernel_size=3,padding=1),#100*64*8*8

ReLU(),

MaxPool2d(2),#100*64*4*4

Flatten(),

Linear(1024,50),#64*4*4

ReLU(),

Dropout(0.5),

Linear(50,10),

Dropout(0.5),

Softmax(-1)

)

def forward(self,x):

x=self.model(x)

return x

pass

if __name__=='__main__':

shen=Shen()

input=torch.ones((100,1,28,28))

output=shen(input)

print(output.shape)

#创建网络模型

shen=Shen()

shen.to(device)#GPU

#损失函数

loss_function=nn.CrossEntropyLoss()

loss_function.to(device)#GPU

#优化器

learning_rate=0.0005

optimizer=torch.optim.Adam(shen.parameters(),lr=learning_rate,betas=(0.9,0.99),eps=1e-8)

#设置网络参数

total_train_step=0

total_test_step=0

epoch=20

#可视化————用Tensorboard --logdir=logs_number --port=6007打开

writer=SummaryWriter('logs_number')

for i in range(epoch):

print('——————第{}轮训练开始——————'.format(i+1))

#训练步骤

shen.train()

for data in train_dataloader:

imgs,targets=data

imgs = imgs.to(device)#GPU

targets = targets.to(device)#GPU

output=shen(imgs)

loss=loss_function(output,targets)

#优化器

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)

pass

pass

#测试步骤

shen.eval()

total_test_loss=0

total_accuracy=0

with torch.no_grad():

for data in test_dataloader:

imgs,targets=data

imgs = imgs.to(device) # GPU

targets = targets.to(device) # GPU

output=shen(imgs)

loss=loss_function(output,targets)

total_test_loss += loss.item()

accuracy=(output.argmax(1)==targets).sum()

total_accuracy += accuracy

pass

pass

print('测试集上的Loss:{}'.format(total_test_loss))

print('测试集上的准确率：{:.4f}%'.format(total_accuracy/test_data_size*100))

writer.add_scalar('test_loss',total_test_loss,total_test_step)

writer.add_scalar('test_accuracy',total_accuracy/test_data_size,total_test_step)

total_test_step += 1

torch.save(shen,'shen_{}.pth'.format(i))

print('模型已保存！')

pass

writer.close()

#斋藤康毅所写的《深度学习入门》中的网络构架与b站up主小土堆的视频讲解的pytorch方法结合进行的一次练手

#所用数据集为MNIST

#困惑：发生了过拟合现象，与理论上99.38%的精度不符，最高只有98.60%的精度，loss函数到一定程度后不再下降