深度學習筆記40_中文文本分類-Pytorch實現

🍨 本文為🔗365天深度學習訓練營?中的學習記錄博客
🍖 原作者：K同學啊 | 接輔導、項目定制

一、我的環境

1.語言環境：Python 3.8

2.編譯器：Pycharm

3.深度學習環境：

torch==1.12.1+cu113
torchvision==0.13.1+cu113

二、導入數據

import torch
import torch.nn as nn
import torchvision
from torchvision import transforms, datasets
import os,PIL,pathlib,warningswarnings.filterwarnings("ignore")             #忽略警告信息
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")import pandas as pd# 加載自定義中文數據
train_data = pd.read_csv('./data/train.csv', sep='\t', header=None)
print(train_data.head())

結果：

                       0              1
0      還有雙鴨山到淮陰的汽車票嗎13號的   Travel-Query
1                從這里怎么回家   Travel-Query
2       隨便播放一首專輯閣樓里的佛里的歌     Music-Play
3              給看一下墓王之王嘛  FilmTele-Play
4  我想看挑戰兩把s686打突變團競的游戲視頻     Video-Play

三、構建詞典

from torchtext.data.utils import get_tokenizer
from torchtext.vocab import build_vocab_from_iterator
import jieba# 中文分詞方法
tokenizer = jieba.lcutdef 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>"]) # 設置默認索引，如果找不到單詞，則會選擇默認索引print(vocab(['我','想','看','和平','精英','上','戰神','必備','技巧','的','游戲','視頻']))

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

text_pipeline  = lambda x: vocab(tokenizer(x))
label_pipeline = lambda x: label_name.index(x)print(text_pipeline('我想看和平精英上戰神必備技巧的游戲視頻'))
print(label_pipeline('Video-Play'))

結果：[2, 10, 13, 973, 1079, 146, 7724, 7574, 7793, 1, 186, 28]
4

四、生成數據批次和迭代器

from torch.utils.data import DataLoaderdef collate_batch(batch):label_list, text_list, offsets = [], [], [0]for (_text, _label) in batch:# 標簽列表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，即語句的總詞匯量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)  # 返回維度dim中輸入元素的累計和return text_list.to(device), label_list.to(device), offsets.to(device)# 數據加載器，調用示例
dataloader = DataLoader(train_iter,batch_size=8,shuffle=False,collate_fn=collate_batch)

五、定義模型

from torch import nnclass TextClassificationModel(nn.Module):def __init__(self, vocab_size, embed_dim, num_class):super(TextClassificationModel, self).__init__()self.embedding = nn.EmbeddingBag(vocab_size,  # 詞典大小embed_dim,  # 嵌入的維度sparse=False)  #self.fc = nn.Linear(embed_dim, num_class)self.init_weights()def init_weights(self):initrange = 0.5self.embedding.weight.data.uniform_(-initrange, initrange)  # 初始化權重self.fc.weight.data.uniform_(-initrange, initrange)self.fc.bias.data.zero_()  # 偏置值歸零def forward(self, text, offsets):embedded = self.embedding(text, offsets)return self.fc(embedded)

六、定義實例

num_class  = len(label_name)
vocab_size = len(vocab)
em_size    = 64
model      = TextClassificationModel(vocab_size, em_size, num_class).to(device)

七、定義訓練函數與評估函數

import timedef train(dataloader):model.train()  # 切換為訓練模式total_acc, train_loss, total_count = 0, 0, 0log_interval = 50start_time = time.time()for idx, (text, label, offsets) in enumerate(dataloader):predicted_label = model(text, offsets)optimizer.zero_grad()  # grad屬性歸零loss = criterion(predicted_label, label)  # 計算網絡輸出和真實值之間的差距，label為真實值loss.backward()  # 反向傳播torch.nn.utils.clip_grad_norm_(model.parameters(), 0.1)  # 梯度裁剪optimizer.step()  # 每一步自動更新# 記錄acc與losstotal_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_timeprint('| epoch {:1d} | {: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, 0start_time = time.time()def evaluate(dataloader):model.eval()  # 切換為測試模式total_acc, train_loss, total_count = 0, 0, 0with torch.no_grad():for idx, (text, label, offsets) in enumerate(dataloader):predicted_label = model(text, offsets)loss = criterion(predicted_label, label)  # 計算loss值# 記錄測試數據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

八、訓練模型

from torch.utils.data.dataset import random_split
from torchtext.data.functional import to_map_style_dataset# 超參數
EPOCHS = 10  # epoch
LR = 5  # 學習率
BATCH_SIZE = 64  # batch size for trainingcriterion = 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# 構建數據集
train_iter = coustom_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)# 獲取當前的學習率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_accprint('-' * 69)print('| epoch {:1d} | time: {:4.2f}s | ''valid_acc {:4.3f} valid_loss {:4.3f} | lr {:4.6f}'.format(epoch,time.time() - epoch_start_time,val_acc, val_loss, lr))print('-' * 69)

?結果：

Batch [50/152], Loss: 0.0340, Accuracy: 0.4203
Batch [100/152], Loss: 0.0235, Accuracy: 0.5851
Batch [150/152], Loss: 0.0309, Accuracy: 0.6572
---------------------------------------------------------------------
| epoch 1 | time: 0.55s | valid_acc 0.814 valid_loss 0.012 | lr 5.000000
---------------------------------------------------------------------
Batch [50/152], Loss: 0.0104, Accuracy: 0.8165
Batch [100/152], Loss: 0.0099, Accuracy: 0.8215
Batch [150/152], Loss: 0.0092, Accuracy: 0.8329
---------------------------------------------------------------------
| epoch 2 | time: 0.44s | valid_acc 0.855 valid_loss 0.008 | lr 5.000000
---------------------------------------------------------------------
Batch [50/152], Loss: 0.0068, Accuracy: 0.8790
Batch [100/152], Loss: 0.0065, Accuracy: 0.8778
Batch [150/152], Loss: 0.0064, Accuracy: 0.8809
---------------------------------------------------------------------
| epoch 3 | time: 0.44s | valid_acc 0.874 valid_loss 0.007 | lr 5.000000
---------------------------------------------------------------------
Batch [50/152], Loss: 0.0050, Accuracy: 0.9105
Batch [100/152], Loss: 0.0051, Accuracy: 0.9101
Batch [150/152], Loss: 0.0048, Accuracy: 0.9130
---------------------------------------------------------------------
| epoch 4 | time: 0.44s | valid_acc 0.882 valid_loss 0.006 | lr 5.000000
---------------------------------------------------------------------
Batch [50/152], Loss: 0.0039, Accuracy: 0.9366
Batch [100/152], Loss: 0.0039, Accuracy: 0.9339
Batch [150/152], Loss: 0.0038, Accuracy: 0.9350
---------------------------------------------------------------------
| epoch 5 | time: 0.44s | valid_acc 0.896 valid_loss 0.006 | lr 5.000000
---------------------------------------------------------------------
Batch [50/152], Loss: 0.0028, Accuracy: 0.9519
Batch [100/152], Loss: 0.0030, Accuracy: 0.9517
Batch [150/152], Loss: 0.0030, Accuracy: 0.9494
---------------------------------------------------------------------
| epoch 6 | time: 0.44s | valid_acc 0.898 valid_loss 0.005 | lr 5.000000
---------------------------------------------------------------------
Batch [50/152], Loss: 0.0025, Accuracy: 0.9580
Batch [100/152], Loss: 0.0024, Accuracy: 0.9616
Batch [150/152], Loss: 0.0024, Accuracy: 0.9609
---------------------------------------------------------------------
| epoch 7 | time: 0.44s | valid_acc 0.902 valid_loss 0.005 | lr 5.000000
---------------------------------------------------------------------
Batch [50/152], Loss: 0.0018, Accuracy: 0.9764
Batch [100/152], Loss: 0.0019, Accuracy: 0.9739
Batch [150/152], Loss: 0.0019, Accuracy: 0.9724
---------------------------------------------------------------------
| epoch 8 | time: 0.44s | valid_acc 0.900 valid_loss 0.005 | lr 5.000000
---------------------------------------------------------------------
Batch [50/152], Loss: 0.0015, Accuracy: 0.9810
Batch [100/152], Loss: 0.0014, Accuracy: 0.9817
Batch [150/152], Loss: 0.0014, Accuracy: 0.9818
---------------------------------------------------------------------
| epoch 9 | time: 0.49s | valid_acc 0.906 valid_loss 0.005 | lr 0.500000
---------------------------------------------------------------------
Batch [50/152], Loss: 0.0013, Accuracy: 0.9831
Batch [100/152], Loss: 0.0013, Accuracy: 0.9831
Batch [150/152], Loss: 0.0014, Accuracy: 0.9825
---------------------------------------------------------------------
| epoch 10 | time: 0.54s | valid_acc 0.906 valid_loss 0.005 | lr 0.500000
---------------------------------------------------------------------

九、預測

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 = "隨便播放一首專輯閣樓里的佛里的歌"
ex_text_str = "還有雙鴨山到淮陰的汽車票嗎13號的"model = model.to("cpu")print("該文本的類別是：%s" %label_name[predict(ex_text_str, text_pipeline)])

該文本的類別是：Travel-Query

總結：?

?語料庫（原始文本）?：
來源包括維基百科、網頁文本、新聞資訊及內部文本。
?文本清洗?：
清洗原始文本，包括去除標點符號和特殊字符。該流程主要用于將原始文本數據轉化為可用于模型訓練的數值化向量，再通過深度學習模型進行文本分類。
- ?分詞?：
  使用jieba分詞工具對清洗后的文本進行分詞處理。
- ?建模?：
  采用不同的模型進行文本建模，包括循環神經網絡（RNN）、卷積神經網絡（CNN）、門控循環單元（GRU）和長短期記憶網絡（LSTM）。
- ?文本向量化?：
  將分詞后的文本轉換為向量表示，方法包括TF-IDF和Word2vec。