Pytorch全连接神经网络实现手写数字识别

左手的ㄟ右手 2024-03-23 12:02 96阅读 0赞

**问题**

Mnist手写数字识别数据集作为一个常见数据集，包含10个类别，在此次深度学习的过程中，我们通过pytorch提供的库函数，运用全连接神经网络实现手写数字的识别

**方法**

设置参数

<table> 
 <tbody> 
  <tr> 
   <td>input_size = 784<br>hidden_size = 500<br>output_size = 10<br>num_epochs = 5<br>batch_size = 100<br>l2earning_rate = 0.001</td> 
  </tr> 
 </tbody> 
</table>

下载mnist数据集，并将其分为训练集和测试集

定义一个带有隐藏层的全连接神经网络

<table> 
 <tbody> 
  <tr> 
   <td>class NeuralNet(nn.Module):<br> &nbsp; &nbsp; def__init__(self,input_size,hidden_size,output_size):<br> &nbsp; &nbsp; &nbsp; &nbsp;super(NeuralNet, self).__init__()<br> &nbsp; &nbsp; &nbsp; &nbsp;self.fc1 = nn.Linear(input_size, hidden_size) <br> &nbsp; &nbsp; &nbsp; &nbsp;self.relu = nn.ReLU()<br> &nbsp; &nbsp; &nbsp; &nbsp;self.fc2 = nn.Linear(hidden_size, output_size) <br>def forward(self, x):<br> &nbsp; &nbsp; out = self.fc1(x)<br> &nbsp; &nbsp; out = self.relu(out)<br> &nbsp; &nbsp; out = self.fc2(out)<br> &nbsp; &nbsp; return out<br>model=NeuralNet(input_size,hidden_size,output_size).to(device) &nbsp; #类的实例化</td> 
  </tr> 
 </tbody> 
</table>

损失函数和优化算法

训练模型

<table> 
 <tbody> 
  <tr> 
   <td>total_step = len(train_loader) &nbsp;#训练数据的大小，也就是含有多少个barch<br>for epoch in range(num_epochs):<br> &nbsp; &nbsp; for i, (images, labels) in enumerate(train_loader): &nbsp; <br> &nbsp; &nbsp; &nbsp; &nbsp;images = images.reshape(-1, 28*28).to(device) &nbsp; &nbsp;<br> &nbsp; &nbsp; &nbsp; &nbsp;labels = labels.to(device) <br> &nbsp; &nbsp; &nbsp; &nbsp;outputs = model(images)<br> &nbsp; &nbsp; &nbsp; &nbsp;loss = criterion(outputs, labels) <br> &nbsp; &nbsp; &nbsp; &nbsp;optimizer.zero_grad()<br> &nbsp; &nbsp; &nbsp; &nbsp;loss.backward()<br> &nbsp; &nbsp; &nbsp; &nbsp;optimizer.step() <br> &nbsp; &nbsp; &nbsp; &nbsp;if (i+1) % 100 == 0:<br> &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;print ('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}' <br> &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; .format(epoch+1, num_epochs, i+1, total_step, loss.item()))</td> 
  </tr> 
 </tbody> 
</table>

测试模型

实验结果

![6dd092b94e053522bab7869f1e45d561.png][]

**结语**

通过此次试验发现，在训练数据时，传入网络的是一个独立标签，即，我们希望输出的是2，但输出的不是用实数2做标签，而是用一个表示实数2的一个十维向量\[0,0,1,0,0,0,0,0,0,0\]，对于分类问题，这种表示尤为重要。

[6dd092b94e053522bab7869f1e45d561.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/03/23/b068a2bc9f06452fa496468bdc29ba25.png