今天接觸深度學(xué)習(xí)，學(xué)習(xí)了多層神經(jīng)網(wǎng)絡(luò)的實(shí)現(xiàn)，完成了一個(gè)小小的實(shí)戰(zhàn)經(jīng)典的手寫數(shù)字識(shí)別的訓(xùn)練。

數(shù)據(jù)集：mnist

網(wǎng)絡(luò):4層每層由一個(gè)線性層+ReLu函數(shù)構(gòu)成

import numpy as np
import torch
from torchvision.datasets import mnist
import matplotlib.pyplot as plt
from torch.utils.data import DataLoader
#使用pytorch自帶的DataLoader定義一個(gè)數(shù)據(jù)迭代器

from torch import nn
from torch.autograd import Variable

#使用內(nèi)置函數(shù)下載mnist數(shù)據(jù)集
train_set = mnist.MNIST('./data',train=True,download=True)
test_set = mnist.MNIST('./data',train=True,download=True)

def data_tf(x):
    x = np.array(x, dtype='float32') / 255
    x = (x - 0.5) / 0.5 #標(biāo)準(zhǔn)化
    x = x.reshape((-1,)) #拉平
    x = torch.from_numpy(x)
    return x

train_set = mnist.MNIST('./data',train=True,transform=data_tf,download=True)#重新加載數(shù)據(jù)集，申明定義的數(shù)據(jù)變換
test_set = mnist.MNIST('./data',train=True,transform=data_tf,download=True)

train_data = DataLoader(train_set,batch_size=64,shuffle=True)
test_data = DataLoader(test_set,batch_size=128,shuffle=True)
#使用這樣的數(shù)據(jù)迭代器是非常有必要的，如果數(shù)據(jù)量太大，就無法一次將他們?nèi)甲x入內(nèi)存，所以需要使用迭代器，每次生成一個(gè)批次的數(shù)據(jù)
a,a_label = next(iter(train_data))
#打印出一個(gè)批次的數(shù)據(jù)大小
print(a.shape)
print(a_label.shape)

#使用Sequential定義4層神經(jīng)網(wǎng)絡(luò)
net = nn.Sequential(
    nn.Linear(784,400),
    nn.ReLU(),
    nn.Linear(400,200),
    nn.ReLU(),
    nn.Linear(200,100),
    nn.ReLU(),
    nn.Linear(100,10),
    nn.ReLU(),
)
print(net)
#使用交叉熵作為loss函數(shù)
#定義loss函數(shù)
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(net.parameters(),1e-1)#使用隨機(jī)梯度下降法，學(xué)習(xí)率0.1
#開始訓(xùn)練
losses = []
acces = []
eval_losses = []
eval_acces = []

for e in range(20):
    train_loss = 0
    train_acc = 0
    net.train()
    for im, label in train_data:
        im = Variable(im)
        label = Variable(label)
        #前向傳播
        out = net(im)
        loss = criterion(out,label)
        #反向傳播
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        #記錄誤差
        train_loss += loss.item()
        #計(jì)算分類準(zhǔn)確性
        _, pred = out.max(1)
        num_correct = (pred == label).sum().item()
        acc = num_correct / im.shape[0]
        train_acc += acc

    losses.append(train_loss / len(train_data))
    acces.append(train_acc / len(train_data))
    #在測(cè)試集上檢驗(yàn)效果
    eval_loss = 0
    eval_acc = 0
    net.eval()#將模型改為預(yù)測(cè)模式
    for im,label in test_data:
        im = Variable(im)
        label = Variable(label)
        out = net(im)
        loss = criterion(out, label)
        #
        eval_loss += loss.item()
        #
        _, pred = out.max(1)
        num_correct = (pred == label).sum().item()
        acc = num_correct / im.shape[0]
        eval_acc += acc
    eval_losses.append(eval_loss / len(test_data))
    eval_acces.append(eval_acc / len(test_data))
    print('epoch:{}, Train_Loss:{:.6f}, Train_Acc:{:.6f}, Eval_Loss:{:.6f}, Eval_Acc:{:.6f}'
            .format(e, train_loss / len(train_data), train_acc / len(train_data),
                    eval_loss / len(test_data), eval_acc / len(test_data)))

plt.title('train loss')
plt.plot(np.arange(len(losses)),losses)
plt.show()
plt.title('train acc')
plt.plot(np.arange(len(acces)), acces)
plt.show()
plt.title('test loss')
plt.plot(np.arange(len(eval_losses)),eval_losses)
plt.show()
plt.title('test acc')
plt.plot(np.arange(len(eval_acces)),eval_acces)
plt.show()