【NLP练习】中文文本分类-Pytorch实现

中文文本分类-Pytorch实现

• 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
• 🍖 原作者：K同学啊

一、准备工作

1. 任务说明

本次使用Pytorch实现中文文本分类。主要代码与文本分类代码基本一致，不同的是本次任务使用了本地的中文数据，数据示例如下：

2.加载数据

import torch
import torch.nn as nn
import torchvision
from torchvision import transforms,datasets
import os,PIL,pathlib,warnings

warnings.filterwarnings("ignore")   #忽略警告信息

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
device

输出：

device(type='cpu')

import pandas as pd
 #加载自定义中文数据
train_data = pd.read_csv('./train.csv',sep='\t',header = None)
train_data.head()

输出：

#构造数据集迭代器
def coustom_data_iter(texts,labels):
 for x,y in zip(texts,labels):
  yield x,y
 
train_iter =coustom_data_iter(train_data[0].values[:],train_data[1].values[:])

二、数据预处理

#构建词典
from torchtext.data.utils import get_tokenizer
from torchtext.vocab import build_vocab_from_iterator
import jieba

#中文分词方法
tokenizer = jieba.lcut
def yield_tokens(data_iter):
    for text,_ in data_iter:
        yield tokenizer(text)

vocab = build_vocab_from_iterator(yield_tokens(train_iter),
                                 specials=["<unk>"])
vocab.set_default_index(vocab["<unk>"])   #设置默认索引，如果找不到单词，则会选择默认索引
vocab(['我','想','看','和平','精英','上','战神','必备','技巧','的','游戏','视频'])

输出：

[2, 10, 13, 973, 1079, 146, 7724, 7574, 7793, 1, 186, 28]

label_name = list(set(train_data[1].values[:]))
print(label_name)

输出：

['FilmTele-Play', 'Alarm-Update', 'Weather-Query', 'Audio-Play', 'Radio-Listen', 'Travel-Query', 'Music-Play', 'Video-Play', 'HomeAppliance-Control', 'Calendar-Query', 'TVProgram-Play', 'Other']

text_pipeline = lambda x : vocab(tokenizer(x))
label_pipeline = lambda x : label_name.index(x)

print(text_pipeline('我想看和平精英上战神必备技巧的游戏视频'))
print(label_pipeline('Video-Play'))

输出：

[2, 10, 13, 973, 1079, 146, 7724, 7574, 7793, 1, 186, 28]
7

lambda表达式的语法为：lambda arguments: expression
其中arguments是函数的参数，可以有多个参数，用逗号分隔。expression是一个表达式，它定义了函数的返回值。

• text_pipeline函数： 将原始文本数据转换为整数列表，使用了之前构建的vocab词表和tokenizer分词器函数。具体步骤：

1. 接受一个字符串x作为输入
2. 使用tokenizer将其分词
3. 将每个词在vocab词表中的索引放入一个列表返回

• label_pipeline函数： 将原始标签数据转换为整数，它接受一个字符串x作为输入，并使用 label_index.index(x) 方法获取x在label_name列表中的索引作为输出。

2.生成数据批次和迭代器

#生成数据批次和迭代器
from torch.utils.data import DataLoader

def collate_batch(batch):
    label_list, text_list, offsets = [],[],[0]         
    for(_text, _label) in batch:
        #标签列表
        label_list.append(label_pipeline(_label))
        #文本列表
        processed_text = torch.tensor(text_pipeline(_text), dtype=torch.int64)
        text_list.append(processed_text)
        #偏移量
        offsets.append(processed_text.size(0))

    label_list = torch.tensor(label_list,dtype=torch.int64)
    text_list = torch.cat(text_list)
    offsets = torch.tensor(offsets[:-1]).cumsum(dim=0)       #返回维度dim中输入元素的累计和
    return text_list.to(device), label_list.to(device), offsets.to(device)

#数据加载器
dataloader = DataLoader(
    train_iter,
    batch_size = 8,
    shuffle = False,
    collate_fn = collate_batch
)

三、模型构建

1. 搭建模型

#搭建模型
from torch import nn

class TextClassificationModel(nn.Module):
    def __init__(self, vocab_size, embed_dim, num_class):
        super(TextClassificationModel,self).__init__()
        self.embedding = nn.EmbeddingBag(vocab_size,      #词典大小
                                        embed_dim,        # 嵌入的维度
                                        sparse=False)     #
        self.fc = nn.Linear(embed_dim, num_class)
        self.init_weights()

    def init_weights(self):
        initrange = 0.5
        self.embedding.weight.data.uniform_(-initrange, initrange)
        self.fc.weight.data.uniform_(-initrange, initrange)
        self.fc.bias.data.zero_()

    def forward(self, text, offsets):
        embedded = self.embedding(text, offsets)
        return self.fc(embedded)

2. 初始化模型

#初始化模型
#定义实例
num_class = len(label_name)
vocab_size = len(vocab)
em_size = 64
model = TextClassificationModel(vocab_size, em_size, num_class).to(device)

3. 定义训练与评估函数

#定义训练与评估函数
import time

def train(dataloader):
    model.train()          #切换为训练模式
    total_acc, train_loss, total_count = 0,0,0
    log_interval = 50
    start_time = time.time()
    for idx, (text,label, offsets) in enumerate(dataloader):
        predicted_label = model(text, offsets)
        optimizer.zero_grad()                             #grad属性归零
        loss = criterion(predicted_label, label)          #计算网络输出和真实值之间的差距，label为真
        loss.backward()                                   #反向传播
        torch.nn.utils.clip_grad_norm_(model.parameters(),0.1)  #梯度裁剪
        optimizer.step()                                  #每一步自动更新
        
        #记录acc与loss
        total_acc += (predicted_label.argmax(1) == label).sum().item()
        train_loss += loss.item()
        total_count += label.size(0)
        if idx % log_interval == 0 and idx > 0:
            elapsed = time.time() - start_time
            print('|epoch{:d}|{:4d}/{:4d} batches|train_acc{:4.3f} train_loss{:4.5f}'.format(
                epoch,
                idx,
                len(dataloader),
                total_acc/total_count,
                train_loss/total_count))
            total_acc,train_loss,total_count = 0,0,0
            staet_time = time.time()

def evaluate(dataloader):
    model.eval()      #切换为测试模式
    total_acc,train_loss,total_count = 0,0,0
    with torch.no_grad():
        for idx,(text,label,offsets) in enumerate(dataloader):
            predicted_label = model(text, offsets)
            loss = criterion(predicted_label,label)   #计算loss值
            #记录测试数据
            total_acc += (predicted_label.argmax(1) == label).sum().item()
            train_loss += loss.item()
            total_count += label.size(0)
    
    return total_acc/total_count, train_loss/total_count

四、训练模型

1. 拆分数据集并运行模型

#拆分数据集并运行模型
from torch.utils.data.dataset   import random_split
from torchtext.data.functional  import to_map_style_dataset

# 超参数设定
EPOCHS      = 10   #epoch
LR          = 5    #learningRate
BATCH_SIZE  = 64   #batch size for training

#设置损失函数、选择优化器、设置学习率调整函数
criterion   = torch.nn.CrossEntropyLoss()
optimizer   = torch.optim.SGD(model.parameters(), lr = LR)
scheduler   = torch.optim.lr_scheduler.StepLR(optimizer, 1.0, gamma = 0.1)
total_accu  = None

# 构建数据集
train_iter = custom_data_iter(train_data[0].values[:],train_data[1].values[:])
train_dataset   = to_map_style_dataset(train_iter)
split_train_, split_valid_ = random_split(train_dataset,
                                         [int(len(train_dataset)*0.8),int(len(train_dataset)*0.2)])

                                           
train_dataloader    = DataLoader(split_train_, batch_size = BATCH_SIZE, shuffle = True, collate_fn = collate_batch)
valid_dataloader    = DataLoader(split_valid_, batch_size = BATCH_SIZE, shuffle = True, collate_fn = collate_batch)

for epoch in range(1, EPOCHS + 1):
    epoch_start_time = time.time()
    train(train_dataloader)
    val_acc, val_loss = evaluate(valid_dataloader)
    #获取当前的学习率
    lr = optimizer.state_dict()['param_groups'][0]['lr']
    if total_accu is not None and total_accu > val_acc:
        scheduler.step()
    else:
        total_accu = val_acc
    print('-' * 69)
    print('| epoch {:d} | time:{:4.2f}s | valid_acc {:4.3f} valid_loss {:4.3f}'.format(
        epoch,
        time.time() - epoch_start_time,
        val_acc,
        val_loss))
    print('-' * 69)

输出：

['还有双鸭山到淮阴的汽车票吗13号的' '从这里怎么回家' '随便播放一首专辑阁楼里的佛里的歌' ...
 '黎耀祥陈豪邓萃雯畲诗曼陈法拉敖嘉年杨怡马浚伟等到场出席' '百事盖世群星星光演唱会有谁' '下周一视频会议的闹钟帮我开开']
|epoch1|  50/ 152 batches|train_acc0.953 train_loss0.00282
|epoch1| 100/ 152 batches|train_acc0.953 train_loss0.00271
|epoch1| 150/ 152 batches|train_acc0.952 train_loss0.00292
---------------------------------------------------------------------
| epoch 1 | time:5.50s | valid_acc 0.949 valid_loss 0.003
---------------------------------------------------------------------
|epoch2|  50/ 152 batches|train_acc0.961 train_loss0.00231
|epoch2| 100/ 152 batches|train_acc0.967 train_loss0.00204
|epoch2| 150/ 152 batches|train_acc0.963 train_loss0.00228
---------------------------------------------------------------------
| epoch 2 | time:5.06s | valid_acc 0.949 valid_loss 0.003
---------------------------------------------------------------------
|epoch3|  50/ 152 batches|train_acc0.975 train_loss0.00173
|epoch3| 100/ 152 batches|train_acc0.973 train_loss0.00177
|epoch3| 150/ 152 batches|train_acc0.972 train_loss0.00166
---------------------------------------------------------------------
| epoch 3 | time:5.07s | valid_acc 0.948 valid_loss 0.003
---------------------------------------------------------------------
|epoch4|  50/ 152 batches|train_acc0.984 train_loss0.00137
|epoch4| 100/ 152 batches|train_acc0.987 train_loss0.00123
|epoch4| 150/ 152 batches|train_acc0.983 train_loss0.00119
---------------------------------------------------------------------
| epoch 4 | time:5.07s | valid_acc 0.950 valid_loss 0.003
---------------------------------------------------------------------
|epoch5|  50/ 152 batches|train_acc0.985 train_loss0.00125
|epoch5| 100/ 152 batches|train_acc0.987 train_loss0.00119
|epoch5| 150/ 152 batches|train_acc0.986 train_loss0.00120
---------------------------------------------------------------------
| epoch 5 | time:5.03s | valid_acc 0.949 valid_loss 0.003
---------------------------------------------------------------------
|epoch6|  50/ 152 batches|train_acc0.985 train_loss0.00118
|epoch6| 100/ 152 batches|train_acc0.989 train_loss0.00114
|epoch6| 150/ 152 batches|train_acc0.985 train_loss0.00120
---------------------------------------------------------------------
| epoch 6 | time:5.40s | valid_acc 0.949 valid_loss 0.003
---------------------------------------------------------------------
|epoch7|  50/ 152 batches|train_acc0.984 train_loss0.00119
|epoch7| 100/ 152 batches|train_acc0.986 train_loss0.00119
|epoch7| 150/ 152 batches|train_acc0.989 train_loss0.00112
---------------------------------------------------------------------
| epoch 7 | time:5.71s | valid_acc 0.949 valid_loss 0.003
---------------------------------------------------------------------
|epoch8|  50/ 152 batches|train_acc0.985 train_loss0.00115
|epoch8| 100/ 152 batches|train_acc0.986 train_loss0.00128
|epoch8| 150/ 152 batches|train_acc0.989 train_loss0.00107
---------------------------------------------------------------------
| epoch 8 | time:5.22s | valid_acc 0.949 valid_loss 0.003
---------------------------------------------------------------------
|epoch9|  50/ 152 batches|train_acc0.988 train_loss0.00114
|epoch9| 100/ 152 batches|train_acc0.983 train_loss0.00127
|epoch9| 150/ 152 batches|train_acc0.989 train_loss0.00109
---------------------------------------------------------------------
| epoch 9 | time:5.28s | valid_acc 0.949 valid_loss 0.003
---------------------------------------------------------------------
|epoch10|  50/ 152 batches|train_acc0.986 train_loss0.00115
|epoch10| 100/ 152 batches|train_acc0.987 train_loss0.00117
|epoch10| 150/ 152 batches|train_acc0.986 train_loss0.00119
---------------------------------------------------------------------
| epoch 10 | time:5.22s | valid_acc 0.949 valid_loss 0.003
---------------------------------------------------------------------

test_acc,test_loss = evaluate(valid_dataloader)
print('模型准确率为：{:5.4f}'.format(test_acc))

输出：

模型准确率为：0.9492

2. 测试指定数据

#测试指定的数据
def predict(text, text_pipeline):
    with torch.no_grad():
        text = torch.tensor(text_pipeline(text))
        output = model(text, torch.tensor([0]))
        return output.argmax(1).item()

ex_text_str = "还有双鸭山到淮阴的汽车票吗13号的"
model = model.to("cpu")

print("该文本的类别是: %s" %label_name[predict(ex_text_str,text_pipeline)])

输出：

该文本的类别是: Travel-Query

五、总结

训练神经网络时，可使用梯度裁剪的方法来防止梯度爆炸，使得模型训练更加稳定