LSTM中文新闻分类源码详解

一、导包

%config InlineBackend.figure_format = 'retina'
%matplotlib inline

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from matplotlib.font_manager import FontProperties 
fonts = FontProperties(fname = "/Library/Fonts/华文细黑.ttf")
import re   
import string  
import copy   
import time   
from sklearn.metrics import accuracy_score,confusion_matrix   

import torch
from torch import nn
import torch.nn.functional as F
import torch.optim as optim
import torch.utils.data as Data
import jieba
jieba.setLogLevel(jieba.logging.INFO)
from torchtext.legacy import data
from torchtext.vocab import Vectors
#从 PyTorch 的拓展库 torchtext 中导入了 Vectors 类，该类用于处理词向量（word embeddings）

二、读取数据

train_df = pd.read_csv("data/lstm/cnews/cnews.train.txt",sep="\t",
                       header=None,names = ["label","text"])
val_df = pd.read_csv("data/lstm/cnews/cnews.val.txt",sep="\t",
                       header=None,names = ["label","text"])
test_df = pd.read_csv("data/lstm/cnews/cnews.test.txt",sep="\t",
                       header=None,names = ["label","text"])
train_df.head(5)

三、数据预处理

stop_words = pd.read_csv("data/lstm/cnews/中文停用词库.txt",
                         header=None,names = ["text"])

1.分词、去掉停用词和数字、字母转换成小写等

## 对中文文本数据进行预处理，去除一些不需要的字符，分词，去停用词，等操作
def chinese_pre(text_data):
    ## 字母转化为小写,去除数字,
    text_data = text_data.lower()
    text_data = re.sub("\d+", "", text_data)
    ## 分词,使用精确模式
    text_data = list(jieba.cut(text_data,cut_all=False)) 
    ## 去停用词和多余空格
    text_data = [word.strip() for word in text_data if word not in stop_words.text.values]
    ## 处理后的词语使用空格连接为字符串
    text_data = " ".join(text_data)
    return text_data

train_df["cutword"] = train_df.text.apply(chinese_pre)
val_df["cutword"] = val_df.text.apply(chinese_pre)
test_df["cutword"] = test_df.text.apply(chinese_pre)

## 预处理后的结果保存为新的文件
train_df[["label","cutword"]].to_csv("data/lstm/cnews_train.csv",index=False)
val_df[["label","cutword"]].to_csv("data/lstm/cnews_val.csv",index=False)
test_df[["label","cutword"]].to_csv("data/lstm/cnews_test.csv",index=False)

train_df.cutword.head()

train_df = pd.read_csv("data/lstm/cnews_train.csv")
val_df = pd.read_csv("data/lstm/cnews_val.csv")
test_df = pd.read_csv("data/lstm/cnews_test.csv")

2.新闻文本标签数值化

labelMap = {"体育": 0,"娱乐": 1,"家居": 2,"房产": 3,"教育": 4,
            "时尚": 5,"时政": 6,"游戏": 7,"科技": 8,"财经": 9}
train_df["labelcode"] =train_df["label"].map(labelMap)
val_df["labelcode"] =val_df["label"].map(labelMap)
test_df["labelcode"] =test_df["label"].map(labelMap)
train_df.head()

train_df[["labelcode","cutword"]].to_csv("data/lstm/cnews_train2.csv",index=False)
val_df[["labelcode","cutword"]].to_csv("data/lstm/cnews_val2.csv",index=False)
test_df[["labelcode","cutword"]].to_csv("data/lstm/cnews_test2.csv",index=False)

三、创建词汇表/词典

1.data.Field()

data.Field参数与方法详解

2.空格切分等

按照空格进行分词，cutword是序列数据，labelcode不是序列数据

## 使用torchtext库进行数据准备
# 定义文件中对文本和标签所要做的操作
"""
sequential=True:表明输入的是序列数据
tokenize="spacy":使用spacy切分词语
use_vocab=True: 创建词汇表
batch_first=True: batch优先的数据方式
fix_length=400 :每个句子固定长度为400，不足会默认使用 <pad> 符号填充
"""
## 定义文本切分方法，因为前面已经做过处理，所以直接使用空格切分即可
mytokenize = lambda x: x.split()
TEXT = data.Field(sequential=True, tokenize=mytokenize, 
                  include_lengths=True, use_vocab=True,
                  batch_first=True, fix_length=400)
LABEL = data.Field(sequential=False, use_vocab=False, 
                   pad_token=None, unk_token=None)
## 对所要读取的数据集的列进行处理
text_data_fields = [
    ("labelcode", LABEL), # 对标签的操作
    ("cutword", TEXT) # 对文本的操作
]
## 读取数据
traindata,valdata,testdata = data.TabularDataset.splits(
    path="data/lstm", format="csv", 
    train="cnews_train2.csv", fields=text_data_fields, 
    validation="cnews_val2.csv",
    test = "cnews_test2.csv", skip_header=True
)

len(traindata),len(valdata),len(testdata)

## 检查一个样本的标签和文本
em = traindata.examples[0]
print(em.labelcode)
print(em.cutword)

3.构建词汇表/词典

使用训练集构建单词表，vectors=None:没有使用预训练好的词向量,而是使用的是随机初始化的词向量，默认是100维

TEXT.build_vocab(traindata,max_size=20000,vectors = None)
LABEL.build_vocab(traindata)
## 可视化训练集中的前50个高频词
word_fre = TEXT.vocab.freqs.most_common(n=50)
word_fre = pd.DataFrame(data=word_fre,columns=["word","fre"])
word_fre.plot(x="word", y="fre", kind="bar",legend=False,figsize=(12,7))
plt.xticks(rotation = 90,fontproperties = fonts,size = 10)
plt.show()

print("词典的词数:",len(TEXT.vocab.itos))
print("前10个单词:\n",TEXT.vocab.itos[0:10])
## 类别标签的数量和类别
print("类别标签情况:",LABEL.vocab.freqs)

这里面的20002，多的那两个应该是

四、构造数据集迭代器，方便批处理

## 定义一个迭代器，将类似长度的示例一起批处理。
BATCH_SIZE = 64
train_iter = data.BucketIterator(traindata,batch_size = BATCH_SIZE)
val_iter = data.BucketIterator(valdata,batch_size = BATCH_SIZE)
test_iter = data.BucketIterator(testdata,batch_size = BATCH_SIZE)

##  获得一个batch的数据，对数据进行内容进行介绍
for step, batch in enumerate(train_iter):  
    if step > 0:
        break
## 针对一个batch 的数据，可以使用batch.labelcode获得数据的类别标签
print("数据的类别标签:\n",batch.labelcode)
## batch.cutword[0]是文本对应的标签向量
print("数据的尺寸:",batch.cutword[0].shape)
## batch.cutword[1] 对应每个batch使用的原始数据中的索引
print("数据样本数:",len(batch.cutword[1]))

batch.cutword[0]和batch.cutword[1]

batch.cutword[0]：表示的是一批数据也就是64条新闻，每条新闻都会被分词，分成一个一个的词语，每个词语在词典中的索引，最后面的1表示的是不足400,填充的对应在词典中的索引为1。

batch.cutword[1]：表示的是一批数据也就是64条新闻，每条新闻对应所有新闻中的索引号。

##  获得一个batch的数据，对数据进行内容进行介绍
for step, batch in enumerate(train_iter):  
    textdata,target = batch.cutword[0],batch.labelcode.view(-1)
    if step > 0:
        break
# ## 针对一个batch 的数据，可以使用batch.labelcode获得数据的类别标签
# print("数据的类别标签:\n",batch.labelcode)
# ## batch.cutword[0]是文本对应的标签向量
# print("数据的尺寸:",batch.cutword[0].shape)
# ## batch.cutword[1] 对应每个batch使用的原始数据中的索引
# print("数据样本数:",len(batch.cutword[1]))

五、搭建LSTM网络

class LSTMNet(nn.Module):
    def __init__(self, vocab_size,embedding_dim, hidden_dim, layer_dim, output_dim):
        """
        vocab_size:词典长度
        embedding_dim:词向量的维度
        hidden_dim: RNN神经元个数
        layer_dim: RNN的层数
        output_dim:隐藏层输出的维度(分类的数量)
        """
        super(LSTMNet, self).__init__()
        self.hidden_dim = hidden_dim ## RNN神经元个数
        self.layer_dim = layer_dim ## RNN的层数
        ## 对文本进行词向量处理
        self.embedding = nn.Embedding(vocab_size, embedding_dim)
        # LSTM ＋ 全连接层
        self.lstm = nn.LSTM(embedding_dim, hidden_dim, layer_dim,
                            batch_first=True)
        self.fc1 = nn.Linear(hidden_dim, output_dim)
    def forward(self, x):
        embeds = self.embedding(x)
        # r_out shape (batch, time_step, output_size)
        # h_n shape (n_layers, batch, hidden_size)   LSTM 有两个 hidden states, h_n 是分线, h_c 是主线
        # h_c shape (n_layers, batch, hidden_size)
        r_out, (h_n, h_c) = self.lstm(embeds, None)   # None 表示 hidden state 会用全0的 state
        # 选取最后一个时间点的out输出
        out = self.fc1(r_out[:, -1, :]) 
        return out
   

r_out, (h_n, h_c)分别是：

r_out是最终输出结果y（根据今天，昨天和日记）

h_n是隐藏层的输出结果s（根据昨天）

h_c是长期信息的输出结果c（根据日记）

vocab_size = len(TEXT.vocab)
embedding_dim = 100
hidden_dim = 128
layer_dim = 1
output_dim = 10
lstmmodel = LSTMNet(vocab_size, embedding_dim, hidden_dim, layer_dim, output_dim)
lstmmodel

六、LSTM网络的训练

## 定义网络的训练过程函数
def train_model2(model,traindataloader, valdataloader,criterion, 
                 optimizer,num_epochs=25,):
    """
    model:网络模型；traindataloader:训练数据集;
    valdataloader:验证数据集，;criterion：损失函数；optimizer：优化方法；
    num_epochs:训练的轮数
    """
    train_loss_all = []
    train_acc_all = []
    val_loss_all = []
    val_acc_all = []
    since = time.time()
    for epoch in range(num_epochs):
        print('-' * 10)
        print('Epoch {}/{}'.format(epoch, num_epochs - 1))
        # 每个epoch有两个阶段,训练阶段和验证阶段
        train_loss = 0.0
        train_corrects = 0
        train_num = 0
        val_loss = 0.0
        val_corrects = 0
        val_num = 0
        model.train() ## 设置模型为训练模式
        for step,batch in enumerate(traindataloader):
            textdata,target = batch.cutword[0],batch.labelcode.view(-1)
            out = model(textdata)
            pre_lab = torch.argmax(out,1) # 预测的标签
            loss = criterion(out, target) # 计算损失函数值
            optimizer.zero_grad()   #梯度清零     
            loss.backward()       #损失函数反向传播
            optimizer.step()        #更新梯度
            train_loss += loss.item() * len(target)
            train_corrects += torch.sum(pre_lab == target.data)
            train_num += len(target)
        ## 计算一个epoch在训练集上的损失和精度
        train_loss_all.append(train_loss / train_num)
        train_acc_all.append(train_corrects.double().item()/train_num)
        print('{} Train Loss: {:.4f}  Train Acc: {:.4f}'.format(
            epoch, train_loss_all[-1], train_acc_all[-1]))
        
        ## 计算一个epoch的训练后在验证集上的损失和精度
        model.eval() ## 设置模型为训练模式评估模式 
        for step,batch in enumerate(valdataloader):
            textdata,target = batch.cutword[0],batch.labelcode.view(-1)
            out = model(textdata)
            pre_lab = torch.argmax(out,1)
            loss = criterion(out, target)   
            val_loss += loss.item() * len(target)
            val_corrects += torch.sum(pre_lab == target.data)
            val_num += len(target)
        ## 计算一个epoch在训练集上的损失和精度
        val_loss_all.append(val_loss / val_num)
        val_acc_all.append(val_corrects.double().item()/val_num)
        print('{} Val Loss: {:.4f}  Val Acc: {:.4f}'.format(
            epoch, val_loss_all[-1], val_acc_all[-1]))
    train_process = pd.DataFrame(
        data={"epoch":range(num_epochs),
              "train_loss_all":train_loss_all,
              "train_acc_all":train_acc_all,
              "val_loss_all":val_loss_all,
              "val_acc_all":val_acc_all})  
    return model,train_process

# 定义优化器
optimizer = torch.optim.Adam(lstmmodel.parameters(), lr=0.0003)  
loss_func = nn.CrossEntropyLoss()   # 损失函数
## 对模型进行迭代训练,对所有的数据训练EPOCH轮
lstmmodel,train_process = train_model2(
    lstmmodel,train_iter,val_iter,loss_func,optimizer,num_epochs=20)

## 输出结果保存和数据保存
torch.save(lstmmodel,"data/lstm/lstmmodel.pkl")
## 导入保存的模型
lstmmodel = torch.load("data/lstm/lstmmodel.pkl")
lstmmodel
## 保存训练过程
train_process.to_csv("data/lstm/lstmmodel_process.csv",index=False)
train_process

## 可视化模型训练过程中
plt.figure(figsize=(18,6))
plt.subplot(1,2,1)
plt.plot(train_process.epoch,train_process.train_loss_all,
         "r.-",label = "Train loss")
plt.plot(train_process.epoch,train_process.val_loss_all,
         "bs-",label = "Val loss")
plt.legend()
plt.xlabel("Epoch number",size = 13)
plt.ylabel("Loss value",size = 13)
plt.subplot(1,2,2)
plt.plot(train_process.epoch,train_process.train_acc_all,
         "r.-",label = "Train acc")
plt.plot(train_process.epoch,train_process.val_acc_all,
         "bs-",label = "Val acc")
plt.xlabel("Epoch number",size = 13)
plt.ylabel("Acc",size = 13)
plt.legend()
plt.show()

七、LSTM网络的测试

## 对测试集进行预测并计算精度
lstmmodel.eval() ## 设置模型为训练模式评估模式 
test_y_all = torch.LongTensor()
pre_lab_all = torch.LongTensor()
for step,batch in enumerate(test_iter):
    textdata,target = batch.cutword[0],batch.labelcode.view(-1)
    out = lstmmodel(textdata)
    pre_lab = torch.argmax(out,1)
    test_y_all = torch.cat((test_y_all,target)) ##测试集的标签
    pre_lab_all = torch.cat((pre_lab_all,pre_lab))##测试集的预测标签

acc = accuracy_score(test_y_all,pre_lab_all)
print("在测试集上的预测精度为:",acc)
## 计算混淆矩阵并可视化
class_label = ["体育","娱乐","家居","房产","教育",
               "时尚","时政","游戏","科技","财经"]
conf_mat = confusion_matrix(test_y_all,pre_lab_all)
df_cm = pd.DataFrame(conf_mat, index=class_label, columns=class_label)
heatmap = sns.heatmap(df_cm, annot=True, fmt="d",cmap="YlGnBu")
heatmap.yaxis.set_ticklabels(heatmap.yaxis.get_ticklabels(), rotation=0,
                             ha='right',fontproperties = fonts)
heatmap.xaxis.set_ticklabels(heatmap.xaxis.get_ticklabels(), rotation=45,
                             ha='right',fontproperties = fonts)
plt.ylabel('True label')
plt.xlabel('Predicted label')
plt.show()