多分類任務中不同隱藏層層數對實驗結果的影響

1 導入實驗所需要的包

import torch
import torch.nn as nn
import numpy as np
import torchvision
import torchvision.transforms as transforms
import matplotlib.pyplot as plt
from torch.utils.data import DataLoader,TensorDataset

2 下載MNIST數據集和讀取數據

train_dataset = torchvision.datasets.MNIST(root="../Datasets/MNIST", train=True, transform=transforms.ToTensor(), download=True)
test_dataset = torchvision.datasets.MNIST(root="../Datasets/MNIST", train=False, transform=transforms.ToTensor(), download=True)
train_x = train_dataset.data.cuda().float() / 255
train_y = train_dataset.targets.cuda().long()
test_x = test_dataset.data.cuda().float() / 255
test_y = test_dataset.targets.cuda().long()
train_dataset = TensorDataset(train_x,train_y)
test_dataset = TensorDataset(test_x,test_y)
batch_size=32
train_iter = DataLoader(train_dataset, batch_size=32, shuffle=True)
test_iter = DataLoader(test_dataset, batch_size=32, shuffle=False)
next(iter(train_iter))[0].shape
next(iter(test_iter))[0].shape

3 定義模型參數

#訓練次數和學習率
num_epochs ,lr = 50, 0.01
num_inputs, num_outputs = 28*28, 10

4 定義模型

第一種：定義一個有 三層 的前饋神經網絡

class LinearNet_1(nn.Module):
    def __init__(self,num_inputs=784, num_outputs=10, num_hiddens=100):
        super(LinearNet_1,self).__init__()
        self.linear1 = nn.Linear(num_inputs,num_hiddens)
        self.relu = nn.ReLU()
        self.linear2 = nn.Linear(num_hiddens,num_outputs)
    
    def forward(self,x):
        x = self.linear1(x)
        x = self.relu(x)
        x = self.linear2(x)
        y = self.relu(x)
        return y

第二種：定義一個有 四層 的前饋神經網絡

class LinearNet_2(nn.Module):
    def __init__(self,num_inputs=784, num_outputs=10, num_hiddens1=100, num_hiddens2=100):
        super(LinearNet_2,self).__init__()
        self.linear1 = nn.Linear(num_inputs,num_hiddens1)
        self.relu = nn.ReLU()
        self.linear2 = nn.Linear(num_hiddens1,num_hiddens2)
        self.linear3 = nn.Linear(num_hiddens2,num_outputs)
    
    def forward(self,x):
        x = self.linear1(x)
        x = self.relu(x)
        x = self.linear2(x)
        x = self.relu(x)
        x = self.linear3(x)
        y = self.relu(x)
        return y

第三種：定義一個有 五層 的前饋神經網絡

class LinearNet_3(nn.Module):
    def __init__(self,num_inputs=784, num_outputs=10, num_hiddens1=100, num_hiddens2=100, num_hiddens3=100):
        super(LinearNet_3,self).__init__()
        self.linear1 = nn.Linear(num_inputs,num_hiddens1)
        self.relu = nn.ReLU()
        self.linear2 = nn.Linear(num_hiddens1,num_hiddens2)
        self.linear3 = nn.Linear(num_hiddens2,num_hiddens3)
        self.linear4 = nn.Linear(num_hiddens3,num_outputs)
    
    def forward(self,x):
        x = self.linear1(x)
        x = self.relu(x)
        x = self.linear2(x)
        x = self.relu(x)
        x = self.linear3(x)
        x = self.relu(x)
        x = self.linear4(x)
        y = self.relu(x)
        return y

5 定義訓練模型

def train(net,train_iter,test_iter,loss,num_epochs,batch_size,params=None,lr=None,optimizer=None):
    train_ls, test_ls = [], []
    for epoch in range(num_epochs):
        ls, count = 0, 0
        for X,y in train_iter:
            X = X.reshape(-1,num_inputs)  #[32, 28, 28]  ->  [32, 784]  
            l=loss(net(X),y)
            optimizer.zero_grad()
            l.backward()
            optimizer.step()
            ls += l.item()*y.shape[0]
        train_ls.append(ls)
        ls, count = 0, 0
        for X,y in test_iter:
            X = X.reshape(-1,num_inputs)
            l=loss(net(X),y)
            ls += l.item()*y.shape[0]
        test_ls.append(ls)
        if(epoch+1)%5==0:
            print('epoch: %d, train loss: %f, test loss: %f'%(epoch+1,train_ls[-1],test_ls[-1]))
    return train_ls,test_ls

6 模型訓練

total_net = [LinearNet_1,LinearNet_2,LinearNet_3]
Train_loss, Test_loss = [], []
#定義損失函數
loss = nn.CrossEntropyLoss()
for cur_net in total_net:
    net = cur_net()
    for param in net.parameters():
        nn.init.normal_(param,mean=0, std= 0.01)
    optimizer = torch.optim.SGD(net.parameters(),lr = 0.001)
    
    train_ls, test_ls = train(net,train_iter,test_iter,loss,num_epochs,batch_size,net.cuda().parameters,lr,optimizer)
    Train_loss.append(train_ls)
    Test_loss.append(test_ls)

7 繪制不同隱藏層數損失圖

x = np.linspace(0,len(train_ls),len(train_ls))
plt.figure(figsize=(10,8))
for i in range(0,3):
    plt.plot(x,Train_loss[i],label= f'with {i+1} hiddens layers:',linewidth=1.5)
    plt.xlabel('epoch')
    plt.ylabel('loss')
plt.legend()
plt.title('train loss')
plt.show()