第N8周：使用Word2vec实现文本分类

🍨 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

本周任务：

结合Word2Vec文本内容预测文本标签

加载数据

python 复制代码

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

warnings.filterwarnings('ignore')

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

python 复制代码

# 从本地CSV文件中读取文本内容和标签
train_data = pd.read_csv("D:/桌面/365/train.csv", sep='\t', header=None)
train_data.head()

python 复制代码

def coustom_data_iter(texts, labels):
    for x, y in zip(texts, labels):
        yield x, y

x = train_data[0].values[:]
y = train_data[1].values[:]

构建词典

python 复制代码

from gensim.models.word2vec import Word2Vec
import numpy as np

w2v = Word2Vec(vector_size=100, min_count=3)

w2v.build_vocab(x)
w2v.train(x, total_examples=w2v.corpus_count, epochs=20)

python 复制代码

def average_vec(text):
    vec = np.zeros(100).reshape((1,100))
    for word in text:
        try:
            vec += w2v.wv[word].reshape((1,100))
        except KeyError:
            continue
    return vec

x_vec = np.concatenate([average_vec(z) for z in x])

w2v.save('./w2c_model.pkl')

生成数据批次和迭代器

python 复制代码

text_pipeline = lambda x: average_vec(x)
label_pipeline = lambda x: label_name.index(x)

python 复制代码

text_pipeline('你在干嘛')

python 复制代码

array([[ 0.78121352,  1.93111382,  0.96291968,  0.39362412, -1.67714586,
        -0.55152619,  1.7284598 ,  0.69204517,  1.1396839 , -0.9755076 ,
        -0.55864345, -3.68676656,  1.41707338, -0.44626126,  0.2580443 ,
         1.09325009,  2.28043211, -2.26334408,  3.32311766, -1.24760717,
         2.2325974 , -0.48408172, -0.55063696,  0.36853465, -1.32127168,
        -0.53377433, -1.48909409, -0.5050023 ,  1.42371842, -0.4252875 ,
         2.52355766,  0.60818394, -1.68924798, -0.16912293,  1.26915893,
        -0.4575564 ,  0.02507078,  3.33139969, -2.1995108 ,  0.44307417,
        -0.41596803,  1.39861814, -0.58643346,  0.91654699, -0.08089826,
         0.08773175,  1.51611513, -0.22212304, -3.55333737,  1.93851076,
         0.42497785, -1.47862379, -0.96684674,  1.20408788, -0.86870126,
        -1.12228102,  1.67186388, -1.11024326, -0.18936946,  1.0811481 ,
         1.82965288, -0.78202841,  2.17574303, -1.03871018, -0.51042572,
         0.40746585, -1.70572275,  1.3409467 ,  1.38298857,  1.11757374,
        -0.8333215 ,  0.04856796,  1.43110101, -0.02333559,  0.82732772,
        -0.9469737 , -4.43783602, -0.20290428,  1.04759257, -1.21757071,
        -1.30356295,  0.50049417, -1.87846385,  2.47995635, -2.41918275,
        -1.72291106,  2.65663178, -0.96948189, -1.30033612, -0.37353188,
         0.53420451, -1.99955091,  0.12223354,  1.74861516,  0.99491888,
        -1.43117569,  0.063243  ,  0.84598846, -2.79536995,  0.02697589]])

python 复制代码

from torch.utils.data import DataLoader

def collate_batch(batch):
    label_list, text_list = [],[]

    for (_text, _label) in batch:
        label_list.append(label_pipeline(_label))

        processed_text = torch.tensor(text_pipeline(_text), dtype=torch.float32)
        text_list.append(processed_text)

    label_list = torch.tensor(label_list, dtype=torch.int64)
    text_list  = torch.cat(text_list)

    return text_list.to(device), label_list.to(device)

datalodaer = DataLoader(train_iter, batch_size=8, shuffle=False, collate_fn=collate_batch)

模型构建

python 复制代码

from torch import nn

class TextClassificationModel(nn.Module):

    def __init__(self, num_class):
        super(TextClassificationModel, self).__init__()
        self.fc = nn.Linear(100, num_class)

    def forward(self, text):
        text = text.float()
        return self.fc(text)

python 复制代码

在这组词汇中不匹配的词汇：书

初始化模型

python 复制代码

num_class = len(label_name)
vocab_size = 100000
em_size = 12
model = TextClassificationModel(num_class).to(device)

定义训练及评估函数

python 复制代码

import time

def train(dataloader):
    model.train()
    total_acc, train_loss, total_count = 0,0,0
    log_interval = 50
    start_time = time.time()

    for idx, (text,label) in enumerate(dataloader): # text, label的顺序不能反，否则会报错

        predicted_label = model(text)

        optimizer.zero_grad()
        loss = criterion(predicted_label, label)
        loss.backward()
        torch.nn.utils.clip_grad_norm_(model.parameters(), 0.1)
        optimizer.step()

        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 {: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 = 0,0,0
            start_time = time.time()
            
def evaluate(dataloader):
    model.eval()
    total_acc,train_loss, total_count = 0,0,0

    with torch.no_grad():
        for idx, (text,label) in enumerate(dataloader):
            predicted_label = model(text)

            loss = criterion(predicted_label, label)

            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

训练模型

python 复制代码

from torch.utils.data.dataset import random_split
from torchtext.data.functional import to_map_style_dataset

EPOCHS = 10
LR = 5
BATCH_SIZE = 64

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

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)

    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 {: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)

python 复制代码

| epoch 1 |   50/ 152 batches| train_acc 0.752 train_loss 0.02433
| epoch 1 |  100/ 152 batches| train_acc 0.836 train_loss 0.01740
| epoch 1 |  150/ 152 batches| train_acc 0.831 train_loss 0.01821
---------------------------------------------------------------------
|epoch 1 | time: 2.83s |valid_acc 0.847 valid_loss 0.016 | lr 5.000000
---------------------------------------------------------------------
| epoch 2 |   50/ 152 batches| train_acc 0.843 train_loss 0.01709
| epoch 2 |  100/ 152 batches| train_acc 0.835 train_loss 0.01863
| epoch 2 |  150/ 152 batches| train_acc 0.854 train_loss 0.01577
---------------------------------------------------------------------
|epoch 2 | time: 1.28s |valid_acc 0.852 valid_loss 0.017 | lr 5.000000
---------------------------------------------------------------------
| epoch 3 |   50/ 152 batches| train_acc 0.854 train_loss 0.01663
| epoch 3 |  100/ 152 batches| train_acc 0.855 train_loss 0.01743
| epoch 3 |  150/ 152 batches| train_acc 0.846 train_loss 0.01738
---------------------------------------------------------------------
|epoch 3 | time: 1.34s |valid_acc 0.862 valid_loss 0.017 | lr 5.000000
---------------------------------------------------------------------
| epoch 4 |   50/ 152 batches| train_acc 0.862 train_loss 0.01514
| epoch 4 |  100/ 152 batches| train_acc 0.854 train_loss 0.01638
| epoch 4 |  150/ 152 batches| train_acc 0.854 train_loss 0.01920
---------------------------------------------------------------------
|epoch 4 | time: 1.18s |valid_acc 0.847 valid_loss 0.018 | lr 5.000000
---------------------------------------------------------------------
| epoch 5 |   50/ 152 batches| train_acc 0.898 train_loss 0.00902
| epoch 5 |  100/ 152 batches| train_acc 0.897 train_loss 0.00885
| epoch 5 |  150/ 152 batches| train_acc 0.900 train_loss 0.00893
---------------------------------------------------------------------
|epoch 5 | time: 1.37s |valid_acc 0.879 valid_loss 0.011 | lr 0.500000
---------------------------------------------------------------------
| epoch 6 |   50/ 152 batches| train_acc 0.900 train_loss 0.00788
| epoch 6 |  100/ 152 batches| train_acc 0.904 train_loss 0.00703
| epoch 6 |  150/ 152 batches| train_acc 0.901 train_loss 0.00681
---------------------------------------------------------------------
|epoch 6 | time: 1.33s |valid_acc 0.883 valid_loss 0.010 | lr 0.500000
---------------------------------------------------------------------
| epoch 7 |   50/ 152 batches| train_acc 0.922 train_loss 0.00573
| epoch 7 |  100/ 152 batches| train_acc 0.901 train_loss 0.00728
| epoch 7 |  150/ 152 batches| train_acc 0.894 train_loss 0.00702
---------------------------------------------------------------------
|epoch 7 | time: 1.12s |valid_acc 0.879 valid_loss 0.009 | lr 0.500000
---------------------------------------------------------------------
| epoch 8 |   50/ 152 batches| train_acc 0.908 train_loss 0.00630
| epoch 8 |  100/ 152 batches| train_acc 0.905 train_loss 0.00593
| epoch 8 |  150/ 152 batches| train_acc 0.911 train_loss 0.00526
---------------------------------------------------------------------
|epoch 8 | time: 1.11s |valid_acc 0.881 valid_loss 0.009 | lr 0.050000
---------------------------------------------------------------------
| epoch 9 |   50/ 152 batches| train_acc 0.911 train_loss 0.00580
| epoch 9 |  100/ 152 batches| train_acc 0.905 train_loss 0.00611
| epoch 9 |  150/ 152 batches| train_acc 0.917 train_loss 0.00516
---------------------------------------------------------------------
|epoch 9 | time: 1.12s |valid_acc 0.881 valid_loss 0.009 | lr 0.005000
---------------------------------------------------------------------
| epoch 10 |   50/ 152 batches| train_acc 0.912 train_loss 0.00564
| epoch 10 |  100/ 152 batches| train_acc 0.905 train_loss 0.00575
| epoch 10 |  150/ 152 batches| train_acc 0.916 train_loss 0.00565
---------------------------------------------------------------------
|epoch 10 | time: 1.12s |valid_acc 0.881 valid_loss 0.009 | lr 0.000500
---------------------------------------------------------------------

测试指定数据

python 复制代码

def predict(text, text_pipeline):
    with torch.no_grad():
        text = torch.tensor(text_pipeline(text), dtype=torch.float32)
        print(text.shape)
        output = model(text)
        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)])

python 复制代码

torch.Size([1, 100])
该文本的类别是: Travel-Query

总结

本周是结合前几周的内容，使用Word2Vec进行词嵌入之后，再实现中文文本分类
本次自己的错误：将for idx, (text,label) in enumerate(dataloader): 中的text、label搞反了，导致输入和模型的输出无法匹配，因此花费了很多时间