原文链接
- 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
- 🍦 参考文章：365天深度学习训练营-第N2周：pytorch中文文本分类识别
- 🍖 原作者：K同学啊|接辅导、项目定制
环境介绍
- 语言环境：Python3.9.12
- 编译器：jupyter notebook
- 深度学习环境：pytorch
前置工作
本次中文文本分类的大致内容是一样的,不一样的地方就在于本次使用的是中文的文本分类
文本分类大致过程如下:
设置环境
在工作开始之前,先保证下载了需要使用到的两个包torchtext和portalocker,我是在anaconda prompt当中下载的,下载 命令如下
pip install torchtext
pip install portalocker
本次训练的内容还需格外下载jieba包,专门用来中文的文本划分
设置GPU
选择运行的设备
import torch
import torch.nn as nn
import os,PIL,pathlib,warnings
warnings.filterwarnings("ignore") #忽略警告信息
# win10系统
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
device
device(type=‘cpu’)
加载数据
本次使用的是当地下载的文件,文件的内容需要的可以私信我或者加入深度学习训练营
首先进行输入的导入
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()
# 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[:])
torchtext.datasets.AG_NEWS()是用于加载AG News数据集的TorchText数据集类,主要包含有世界,科技,体育和商业等新闻文章
构建词典
from torchtext.data.utils import get_tokenizer
from torchtext.vocab import build_vocab_from_iterator
# conda install jieba -y
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'])
label_name = list(set(train_data[1].values[:]))
print(label_name)
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'))
[[‘Music-Play’, ‘Alarm-Update’, ‘TVProgram-Play’, ‘FilmTele-Play’, ‘Video-Play’, ‘Weather-Query’, ‘HomeAppliance-Control’, ‘Other’, ‘Audio-Play’, ‘Calendar-Query’, ‘Travel-Query’, ‘Radio-Listen’]
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
4
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，即语句的总词汇量
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，调用示例
dataloader = DataLoader(train_iter,
batch_size=8,
shuffle =False,
collate_fn=collate_batch)
模型定义
首先先定义我们进行分类用到的模型,然后嵌入文本,然后对句子嵌入之后的结果均值聚合
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)
self.embedding.weight.data.uniform_(-initrange, initrange)这段代码在PyTorch框架下用于初始化神经网络的词嵌入层权重的一种方法,这样使得模型在训练时具有一定的随机性,避免了梯度消失或者梯度爆炸等问题
定义实例
在设置好的模型当中进行一个命名为model
num_class = len(set([label for (label, text) in train_iter]))
vocab_size = len(vocab)
em_size = 64
model = TextClassificationModel(vocab_size, em_size, num_class).to(device)
定义训练函数和评估函数
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为真实值
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 {: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() # 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
模型训练
from torch.utils.data.dataset import random_split
from torchtext.data.functional import to_map_style_dataset
# 超参数
EPOCHS = 10 # epoch
LR = 5 # 学习率
BATCH_SIZE = 64 # batch size for training
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 = 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)
# 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 {: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)
| epoch 1 | 50/ 152 batches | train_acc 0.422 train_loss 0.03069
| epoch 1 | 100/ 152 batches | train_acc 0.698 train_loss 0.01958
| epoch 1 | 150/ 152 batches | train_acc 0.765 train_loss 0.01386
---------------------------------------------------------------------
| epoch 1 | time: 1.55s | valid_acc 0.799 valid_loss 0.012 | lr 5.000000
---------------------------------------------------------------------
| epoch 2 | 50/ 152 batches | train_acc 0.811 train_loss 0.01107
| epoch 2 | 100/ 152 batches | train_acc 0.832 train_loss 0.00908
| epoch 2 | 150/ 152 batches | train_acc 0.849 train_loss 0.00813
---------------------------------------------------------------------
| epoch 2 | time: 1.51s | valid_acc 0.849 valid_loss 0.008 | lr 5.000000
---------------------------------------------------------------------
| epoch 3 | 50/ 152 batches | train_acc 0.869 train_loss 0.00679
| epoch 3 | 100/ 152 batches | train_acc 0.886 train_loss 0.00646
| epoch 3 | 150/ 152 batches | train_acc 0.881 train_loss 0.00629
---------------------------------------------------------------------
| epoch 3 | time: 1.49s | valid_acc 0.874 valid_loss 0.007 | lr 5.000000
---------------------------------------------------------------------
| epoch 4 | 50/ 152 batches | train_acc 0.900 train_loss 0.00529
| epoch 4 | 100/ 152 batches | train_acc 0.910 train_loss 0.00488
| epoch 4 | 150/ 152 batches | train_acc 0.915 train_loss 0.00473
---------------------------------------------------------------------
| epoch 4 | time: 1.61s | valid_acc 0.881 valid_loss 0.006 | lr 5.000000
---------------------------------------------------------------------
| epoch 5 | 50/ 152 batches | train_acc 0.931 train_loss 0.00385
...
| epoch 10 | 150/ 152 batches | train_acc 0.983 train_loss 0.00128
---------------------------------------------------------------------
| epoch 10 | time: 1.71s | valid_acc 0.901 valid_loss 0.005 | lr 5.000000
---------------------------------------------------------------------
TorchText是Torch的一个拓展库,专注于处理文本数据,这样我们可以通过索引直接访问数据集当中的特定样本,简化了模型的训练,验证和测试过程当中的数据处理
模型预测
#预测函数
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 = "随便播放一首专辑阁楼里的佛里的歌"
#ex_text_str = "还有双鸭山到淮阴的汽车票吗13号的"
model = model.to(device)
print("该文本的类别是：%s" % label_name[predict(ex_text_str, text_pipeline)])
该文本的类别是：Music-Play
3876
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
