---

项目背景

项目的目的，是为了对情感评论数据集进行预测打标。在训练之前，需要对数据进行数据清洗环节，前面已对数据进行清洗，详情可移步至NLP_情感分类_数据清洗
前面用机器学习方案解决，详情可移步至NLP_情感分类_机器学习方案

下面对已清洗的数据集，用序列模型方案进行处理

代码

导包

import warnings
warnings.filterwarnings('ignore')
import pandas as pd
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder
from torchtext.data.utils import get_tokenizer
from torchtext.vocab import build_vocab_from_iterator
from torch.nn.utils.rnn import pad_sequence
import torch.nn.functional as F  

读取数据

df = pd.read_csv('data/sentiment_analysis_clean.csv')
df = df.dropna()

文本预处理

tokenizer = get_tokenizer('basic_english')
def yield_tokens(data_iter):
    for text in data_iter:
        yield tokenizer(text)

vocab = build_vocab_from_iterator(yield_tokens(df['text']), specials=["<unk>"])
vocab.set_default_index(vocab["<unk>"])

# 标签编码
label_encoder = LabelEncoder()
df['label'] = label_encoder.fit_transform(df['label'])

# 划分训练集和测试集
train_df, test_df = train_test_split(df, test_size=0.2, random_state=2024)

# 转换为 PyTorch 张量
def text_pipeline(x):
    return vocab(tokenizer(x))

def label_pipeline(x):
    return int(x)

举例查看分词器

tokenizer('I like apple'),vocab(tokenizer('I like apple'))

数据集调整

class TextDataset(Dataset):
    def __init__(self, df):
        self.texts = df['text'].values
        self.labels = df['label'].values

    def __len__(self):
        return len(self.texts)

    def __getitem__(self, idx):
        text = torch.tensor(text_pipeline(self.texts[idx]), dtype=torch.long)
        label = torch.tensor(label_pipeline(self.labels[idx]), dtype=torch.long)
        return text, label

train_dataset = TextDataset(train_df)
test_dataset = TextDataset(test_df)

def collate_batch(batch):
    text_list, label_list = [], []
    for (text, label) in batch:
        text_list.append(text)
        label_list.append(label)
    text_list = pad_sequence(text_list, batch_first=True, padding_value=vocab['<pad>'])
    label_list = torch.tensor(label_list, dtype=torch.long)
    return text_list, label_list

train_loader = DataLoader(train_dataset, batch_size=128, shuffle=True, collate_fn=collate_batch)
test_loader = DataLoader(test_dataset, batch_size=128, shuffle=False, collate_fn=collate_batch)

进一步剖析：对应Step [{i+1}/{len(train_loader)}] 里的train_loader

len(train_dataset)

进一步剖析：Step [{i+1}/{len(train_loader)}] 里的train_loader，原始的train_df

len(train_df)

计算数据集中最长文本的长度

max_seq_len = 0
for text in df['text']:
    tokens = text_pipeline(text)
    if len(tokens) > max_seq_len:
        max_seq_len = len(tokens)

print(f'Max sequence length in the dataset: {max_seq_len}')

定义模型

class TextClassifier(nn.Module):
    def __init__(self, model_type, vocab_size, embed_dim, hidden_dim, output_dim, num_layers=1):
        super(TextClassifier, self).__init__()
        self.model_type = model_type
        self.embedding = nn.Embedding(vocab_size, embed_dim, padding_idx=vocab['<pad>'])

        if model_type == 'RNN':
            self.rnn = nn.RNN(embed_dim, hidden_dim, num_layers, batch_first=True)
        elif model_type == 'GRU':
            self.rnn = nn.GRU(embed_dim, hidden_dim, num_layers, batch_first=True)
        elif model_type == 'LSTM':
            self.rnn = nn.LSTM(embed_dim, hidden_dim, num_layers, batch_first=True)
        else:
            raise ValueError("model_type should be one of ['RNN', 'GRU', 'LSTM']")

        if model_type in ['RNN', 'GRU', 'LSTM']:
            self.fc = nn.Linear(hidden_dim, output_dim)
    def forward(self, x):
        x = self.embedding(x) # [batch,seq_len,emb_dim]
        if self.model_type in ['RNN', 'GRU', 'LSTM']:
            h0 = torch.zeros(self.rnn.num_layers, x.size(0), self.rnn.hidden_size).to(x.device)
            if self.model_type == 'LSTM':
                c0 = torch.zeros(self.rnn.num_layers, x.size(0), self.rnn.hidden_size).to(x.device)
                out, _ = self.rnn(x, (h0, c0))
            else:
                out, _ = self.rnn(x, h0)
            # out :[batch,seq_len,emb_dim]
            out = self.fc(out[:, -1, :]) # 使用输出序列最后一个时间步的表征作为序列整体的表征
        else:
            raise ValueError("model_type should be one of ['RNN', 'GRU', 'LSTM']")           
        return out

def train_model(model, train_loader, criterion, optimizer, num_epochs=2, device='cpu'):
    model.to(device)
    model.train()
    for epoch in range(num_epochs):
        running_loss = 0.0
        correct = 0
        total = 0
        for i, (texts, labels) in enumerate(train_loader):
            texts, labels = texts.to(device), labels.to(device)
            outputs = model(texts)
            loss = criterion(outputs, labels)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()

            running_loss += loss.item()
            
            _, predicted = torch.max(outputs.data, 1)
            total += labels.size(0)
            correct += (predicted == labels).sum().item()
            
            if i % 10 == 0:  # 每个批次输出一次日志
                accuracy = 100 * correct / total
                print(f'Epoch [{epoch+1}/{num_epochs}], Step [{i+1}/{len(train_loader)}], Loss: {loss.item():.4f}, Accuracy: {accuracy:.2f}%')
        
        epoch_loss = running_loss / len(train_loader)
        epoch_accuracy = 100 * correct / total
        print(f'Epoch [{epoch+1}/{num_epochs}], Average Loss: {epoch_loss:.4f}, Average Accuracy: {epoch_accuracy:.2f}%')

def evaluate_model(model, test_loader, device='cpu'):
    model.to(device)
    model.eval()
    correct = 0
    total = 0
    with torch.no_grad():
        for texts, labels in test_loader:
            texts, labels = texts.to(device), labels.to(device)
            outputs = model(texts)
            _, predicted = torch.max(outputs.data, 1)
            total += labels.size(0)
            correct += (predicted == labels).sum().item()
    print(f'Accuracy: {100 * correct / total:.2f}%')

超参数

vocab_size = len(vocab)
embed_dim = 128
hidden_dim = 128
output_dim = len(label_encoder.classes_)
num_layers = 1
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

进一步剖析label_encoder.classes_

len(label_encoder.classes_)

训练 RNN 模型

model_rnn = TextClassifier('RNN', vocab_size, embed_dim, hidden_dim, output_dim, num_layers)
criterion = nn.CrossEntropyLoss()
optimizer_rnn = optim.Adam(model_rnn.parameters(), lr=0.001)
train_model(model_rnn, train_loader, criterion, optimizer_rnn, num_epochs=2, device=device)
evaluate_model(model_rnn, test_loader, device=device)

训练 GRU 模型

model_gru = TextClassifier('GRU', vocab_size, embed_dim, hidden_dim, output_dim, num_layers)
optimizer_gru = optim.Adam(model_gru.parameters(), lr=0.001)
train_model(model_gru, train_loader, criterion, optimizer_gru, num_epochs=2, device=device)
evaluate_model(model_gru, test_loader, device=device)

训练 LSTM 模型

model_lstm = TextClassifier('LSTM', vocab_size, embed_dim, hidden_dim, output_dim, num_layers)
optimizer_lstm = optim.Adam(model_lstm.parameters(), lr=0.001)
train_model(model_lstm, train_loader, criterion, optimizer_lstm, num_epochs=2, device=device)
evaluate_model(model_lstm, test_loader, device=device)

同类型项目

阿里云-零基础入门NLP【基于机器学习的文本分类】

阿里云-零基础入门NLP【基于深度学习的文本分类3-BERT】
也可以参考进行学习

---

学习的参考资料：
深度之眼