pytorch 手写数字识别项目 增量式训练

Posted 2021-03-09 liuxinyu12378

dataset.py

 

‘‘‘
准备数据集
‘‘‘
import torch
from torch.utils.data import DataLoader
from torchvision.datasets import MNIST
from torchvision.transforms import ToTensor,Compose,Normalize
import torchvision
import config

def mnist_dataset(train):
    func = torchvision.transforms.Compose([
        torchvision.transforms.ToTensor(),
        torchvision.transforms.Normalize(
            mean=(0.1307,),
            std = (0.3081,)
        )
    ])

    #准备Mnist数据集
    return MNIST(root="../mnist",train=train,download=False,transform=func)

def get_dataloader(train = True):
    mnist = mnist_dataset(train)
    batch_size = config.train_batch_size if train else config.test_batch_size
    return DataLoader(mnist,batch_size=batch_size,shuffle=True)

if __name__ == ‘__main__‘:
    for (images,labels) in get_dataloader():
        print(images.size())
        print(labels)
        break

 

　　model.py

 

‘‘‘定义模型‘‘‘

import torch.nn as nn
import torch.nn.functional as F

class MnistModel(nn.Module):
    def __init__(self):
        super(MnistModel,self).__init__()
        self.fc1 = nn.Linear(28*28,100)
        self.fc2 = nn.Linear(100,10)

    def forward(self,image):
        image_viwed = image.view(-1,28*28)
        fc1_out = self.fc1(image_viwed)
        fc1_out_relu = F.relu(fc1_out)
        out = self.fc2(fc1_out_relu)

        return F.log_softmax(out,dim=-1)

 

　　config.py

 

‘‘‘
项目配置
‘‘‘
import torch

train_batch_size = 128
test_batch_size = 128

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

 

　　train.py

 

‘‘‘
进行模型的训练
‘‘‘
from dataset import get_dataloader
from models import MnistModel
from torch import optim
import torch.nn.functional as F
import config
from tqdm import tqdm
import numpy as np
import torch
import os
from eval import eval

#实例化模型、优化器、损失函数
model = MnistModel().to(config.device)
optimizer = optim.Adam(model.parameters(),lr=0.001)

if os.path.exists("./model/mnist_net.pt"):
    model.load_state_dict(torch.load("./model/mnist_net.pt"))
    optimizer.load_state_dict(torch.load("model/mnist_optimizer.pt"))


#迭代训练

def train(epoch):
    train_dataloader = get_dataloader(train=True)
    bar = tqdm(enumerate(train_dataloader),total=len(train_dataloader))
    total_loss = []
    for idx,(input,target) in bar:
        input = input.to(config.device)
        target = target.to(config.device)
        #梯度置为0
        optimizer.zero_grad()
        #计算得到预测值
        output = model(input)
        #得到损失
        loss = F.nll_loss(output,target)
        total_loss.append(loss.item())
        #反向传播，计算损失
        loss.backward()
        #参数更新
        optimizer.step()

        if idx%10 ==0:
            bar.set_description("epoch:{} idx:{},loss:{}".format(epoch,idx,np.mean(total_loss)))
            torch.save(model.state_dict(),"model/mnist_net.pt")
            torch.save(optimizer.state_dict(),"model/mnist_optimizer.pt")

if __name__ == ‘__main__‘:
    for i in range(10):
        train(i)
        eval()

 

　　eval.py

 

‘‘‘
进行模型的训练
‘‘‘
from dataset import get_dataloader
from models import MnistModel
from torch import optim
import torch.nn.functional as F
import config
import numpy as np
import torch
import os




#迭代训练

def eval():
    # 实例化模型、优化器、损失函数
    model = MnistModel().to(config.device)
    optimizer = optim.Adam(model.parameters(), lr=0.01)

    if os.path.exists("./model/mnist_net.pt"):
        model.load_state_dict(torch.load("./model/mnist_net.pt"))
        optimizer.load_state_dict(torch.load("model/mnist_optimizer.pt"))
    test_dataloader = get_dataloader(train=False)
    total_loss = []
    total_acc = []
    with torch.no_grad():
        for input,target in test_dataloader:
            input = input.to(config.device)
            target = target.to(config.device)
            #计算得到预测值
            output = model(input)
            #计算损失
            loss = F.nll_loss(output,target)
            #反向传播，计算损失
            total_loss.append(loss.item())
            #计算准确率
            pred = output.max(dim=-1)[-1]
            total_acc.append(pred.eq(target).float().mean().item())
    print("test loss:{},test acc:{}".format(np.mean(total_loss),np.mean(total_acc)))

if __name__ == ‘__main__‘:
        eval()

 

　　

D:anacondapython.exe C:/Users/liuxinyu/Desktop/pytorch_test/day3/手写数字识别/train.py
epoch:0 idx:460,loss:0.32289110562095413: 100%|██████████| 469/469 [00:24<00:00, 19.05it/s]
test loss:0.17968503131142147,test acc:0.9453125
epoch:1 idx:460,loss:0.15012750004513145: 100%|█████████▉| 468/469 [00:20<00:00, 22.10it/s]epoch:1 idx:460,loss:0.15012750004513145: 100%|██████████| 469/469 [00:20<00:00, 22.52it/s]
test loss:0.12370304338916947,test acc:0.9624208860759493
epoch:2 idx:460,loss:0.10398845713577534:  99%|█████████▉| 464/469 [00:21<00:00, 22.78it/s]epoch:2 idx:460,loss:0.10398845713577534: 100%|█████████▉| 467/469 [00:21<00:00, 22.71it/s]epoch:2 idx:460,loss:0.10398845713577534: 100%|██████████| 469/469 [00:21<00:00, 21.82it/s]
test loss:0.10385569722592077,test acc:0.9697389240506329
epoch:3 idx:460,loss:0.07973297938720653: 100%|█████████▉| 467/469 [00:22<00:00, 23.12it/s]epoch:3 idx:460,loss:0.07973297938720653: 100%|██████████| 469/469 [00:22<00:00, 20.84it/s]
test loss:0.08691684670652015,test acc:0.9754746835443038
epoch:4 idx:460,loss:0.0650228117158285: 100%|█████████▉| 468/469 [00:21<00:00, 24.06it/s]epoch:4 idx:460,loss:0.0650228117158285: 100%|██████████| 469/469 [00:21<00:00, 21.79it/s]
test loss:0.0803159438309413,test acc:0.9760680379746836
epoch:5 idx:460,loss:0.05270117848966101: 100%|██████████| 469/469 [00:21<00:00, 21.92it/s]
test loss:0.08102699166423158,test acc:0.9759691455696202
epoch:6 idx:460,loss:0.04386751471317642: 100%|██████████| 469/469 [00:19<00:00, 24.58it/s]
test loss:0.07991968260347089,test acc:0.9769580696202531
epoch:7 idx:460,loss:0.03656852366544161: 100%|██████████| 469/469 [00:15<00:00, 31.20it/s]
test loss:0.07767781678917288,test acc:0.9774525316455697
epoch:8 idx:460,loss:0.03112584312896925: 100%|██████████| 469/469 [00:14<00:00, 32.41it/s]
test loss:0.07755146227494071,test acc:0.9773536392405063
epoch:9 idx:460,loss:0.025217091969725495: 100%|██████████| 469/469 [00:14<00:00, 31.53it/s]
test loss:0.07112929566845863,test acc:0.9802215189873418