• 用pytorch做手写数字识别,识别l率达97.8%


    pytorch做手写数字识别

    效果如下:

    工程目录如下

    第一步  数据获取

    下载MNIST库,这个库在网上,执行下面代码自动下载到当前data文件夹下

    from torchvision.datasets import MNIST
    import torchvision
    
    mnist = MNIST(root='./data',train=True,download=True)
    
    print(mnist)
    print(mnist[0])
    print(len(mnist))
    img = mnist[0][0]
    img.show()
    

      

    dataset.py文件,读取数据并做预处理

    '''
    准备数据集
    '''
    
    import torch
    from torch.utils.data import DataLoader
    from torchvision.datasets import MNIST
    import torchvision
    
    
    def mnist_dataset(train):
    
        func = torchvision.transforms.Compose([
            torchvision.transforms.ToTensor(),
            torchvision.transforms.Normalize(mean=(0.1307,),std=(0.3081,))
        ])
    
        #1.准备Mnist数据集
        return MNIST(root='./data',train=train,download=False,transform=func)
    
    def get_dataloader(train = True):
        mnist = mnist_dataset(train)
        return DataLoader(mnist,batch_size=128,shuffle=True)
    
    if __name__ == '__main__':
        for (images,labels) in get_dataloader():
            print(images.size())
            print(labels.size())
            break
    

      

    models.py文件,定义训练的模型类

    '''
    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(1*28*28,100)
            self.fc2 = nn.Linear(100,10)
    
        def forward(self,image):
            image_viewd = image.view(-1,1*28*28) #[batch_size,1*28*28]
            fc1_out = self.fc1(image_viewd) #[batch_size,100]
            fc1_out_relu = F.relu(fc1_out) #[batch_size,100]
            out = self.fc2(fc1_out_relu) #[batch_size,10]
    
            return F.log_softmax(out,dim=-1)  #带权损失计算交叉熵
    

    cong.py文件,定义一些常亮,设置使用cpu还是GPU  

    '''
    项目配置
    '''
    
    import torch
    
    train_batch_size = 128
    test_batch_size = 100
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    

      

    train.py文件,模型训练文件,保存模型

    """
    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 conf
    from tqdm import tqdm
    import numpy as np
    import torch
    import os
    from test import eval
    
    #1. 实例化模型,优化器,损失函数
    model = MnistModel().to(conf.device)
    optimizer = optim.Adam(model.parameters(),lr=1e-3)
    
    #2. 进行循环,进行训练
    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(conf.device)
            target = target.to(conf.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)
            #反向传播,计算损失
            loss.backward()
            total_loss.append(loss.item())
            #参数的更新
            optimizer.step()
            #打印数据
            if idx%10 ==0 :
                bar.set_description_str("epcoh:{} idx:{},loss:{:.6f}".format(epoch,idx,np.mean(total_loss)))
                torch.save(model.state_dict(),"./models/model.pkl")
                torch.save(optimizer.state_dict(),"./models/optimizer.pkl")
    
    if __name__ == '__main__':
        for i in range(10):
            train(i)
            eval()
    

    test.py文件,模型测试文件,测试模型准确率  

    '''
    进行模型评估
    '''
    
    from dataset import get_dataloader
    from models import MnistModel
    from torch import optim
    import torch.nn.functional as F
    import conf
    from tqdm import tqdm
    import numpy as np
    import torch
    import os
    
    def eval():
        #实例化模型,优化器,损失函数
        model = MnistModel().to(conf.device)
    
        if os.path.exists("./models/model.pkl"):
            model.load_state_dict(torch.load("./models/model.pkl"))
    
        test_dataloader = get_dataloader(train=False)
        total_loss = []
        total_acc = []
        with torch.no_grad():
            for input, target in test_dataloader:  # 2. 进行循环,进行训练
                input = input.to(conf.device)
                target = target.to(conf.device)
                # 计算得到预测值
                output = model(input)
                # 得到损失
                loss = F.nll_loss(output, target)
                # 反向传播,计算损失
                total_loss.append(loss.item())
    
                # 计算准确率
                ###计算预测值
                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__':
    #     # for i in range(10):
    #     #     train(i)
    #     eval()
    

      

    多思考也是一种努力,做出正确的分析和选择,因为我们的时间和精力都有限,所以把时间花在更有价值的地方。
  • 相关阅读:
    基于Freescale的主流芯片HCS08
    BizTalk Server 2010 映射器(Mapper) [ 下篇 ]
    BizTalk Server 2010 使用 WCF Service [ 中篇 ]
    Ext JS 4 Beta 1发布了
    Step by Step WebMatrix网站开发之一:Webmatrix安装
    REST WebService与SOAP WebService的比较
    BizTalk Server 2010 使用 WCF Service [ 上篇 ]
    BizTalk Server 2010 映射器(Mapper) [ 中篇 ]
    BizTalk Server 2010 映射器(Mapper) [ 上篇 ]
    ExtJS 4 Beta 2预览:Ext.Brew包
  • 原文地址:https://www.cnblogs.com/LiuXinyu12378/p/11392033.html
Copyright © 2020-2023  润新知