Datawhale-零基础入门NLP-新闻文本分类Task04

1 FastText 学习路径

FastText 是 facebook 近期开源的一个词向量计算以及文本分类工具,FastText的学习路径为：

具体原理就不作解析了,详细教程见：https://fasttext.cc/docs/en/support.html

2 FastText 安装

2.1 基于框架的安装

需要从github下载源码，然后生成可执行的fasttext文件

（1）命令：git clone https://github.com/facebookresearch/fastText.git

（2）命令：cd fastText/  and   ls  

（3）命令：make

2.2 基于Python模块的安装

（1）直接pip安装：pip install fasttext

（2）源码安装：

3 FastText 实现文本分类

3.1 例子

（1）下载数据

#读取数据
wget https://dl.fbaipublicfiles.com/fasttext/data/cooking.stackexchange.tar.gz
#解压数据
tar xvzf cooking.stackexchange.tar.gz
#显示前几行
head cooking.stackexchange.txt

（2）划分数据集

#查看数据
wc cooking.stackexchange.txt

#划分数据集
head -n 12404 cooking.stackexchange.txt > cooking.train
tail -n 3000 cooking.stackexchange.txt > cooking.valid

（3）训练与调参

此处是基于命令行的展示，Python的展示可参考：https://fasttext.cc/docs/en/supervised-tutorial.html

fasttext的参数有：

训练：

./fasttext supervised -input cooking.train -output model_cooking

预测：

 ./fasttext predict model_cooking.bin -

3.2 基于新闻文本的FastText分析

import fasttext
import pandas as pd
from sklearn.metrics import f1_score

train_df = pd.read_csv('data/data45216/train_set.csv',sep='\t')

train_df['label_ft'] = '__label__' + train_df['label'].astype(str) 
train_df[['text','label_ft']].iloc[:-5000].to_csv('train.csv',index=None,header=None,sep='\t')

model = fasttext.train_supervised('train.csv',lr=1.0,wordNgrams=2,verbose=2,minCount=1,epoch=25,loss='hs')

val_pred = [model.predict(x)[0][0].split('__')[-1] for x in  train_df.iloc[-5000:]['text']]
print(f1_score(train_df['label'].values[-5000:].astype(str),val_pred,average='macro'))

输出结果为：

4 FastText调参

FastText的train_supervised参数有：

可通过以上参数进行手动设置，也可用过FastText的自动调参功能进行调参。

4.1 基于命令行

（1）验证集验证-autotune-validation

./fasttext supervised -input cooking.train -output model_cooking -autotune-validation cooking.valid

       （2）设置执行时间-autotune-duration

./fasttext supervised -input cooking.train -output model_cooking -autotune-validation cooking.valid -autotune-duration 600

       （3）模型大小 -autotune-modelsize

./fasttext supervised -input cooking.train -output model_cooking -autotune-validation cooking.valid -autotune-modelsize 2M

（4）指标 -autotune-metric

-autotune-metric f1:__label__baking
-autotune-metric precisionAtRecall:30
-autotune-metric precisionAtRecall:30:__label__baking
-autotune-metric recallAtPrecision:30
-autotune-metric recallAtPrecision:30:__label__baking

4.2 基于Python模块

（1）验证集验证autotuneValidationFile

model = fasttext.train_supervised(input='cooking.train', autotuneValidationFile='cooking.valid')

       （2）设置执行时间autotuneDuration

model = fasttext.train_supervised(input='cooking.train', autotuneValidationFile='cooking.valid', autotuneDuration=600)

       （3）模型大小autotuneModelSize

model = fasttext.train_supervised(input='cooking.train', autotuneValidationFile='cooking.valid', autotuneModelSize="2M")

（4）指标 autotuneMetric

model = fasttext.train_supervised(input='cooking.train', autotuneValidationFile='cooking.valid', autotuneMetric="f1:__label__baking")

5 作业

使用自动调参进行训练：

import fasttext
import pandas as pd
from sklearn.metrics import f1_score

train_df = pd.read_csv('data/data45216/train_set.csv',sep='\t')

#将label值转成fasttext识别的格式
train_df['label_ft'] = '__label__' + train_df['label'].astype(str) 
#划分训练集和验证集
train_df[['text','label_ft']].iloc[:10000].to_csv('train.csv',index=None,header=None,sep='\t')
train_df[['text','label_ft']].iloc[10000:15000].to_csv('valid.csv',index=None,header=None,sep='\t')

#建立模型
model = fasttext.train_supervised('train.csv',lr=1.0,wordNgrams=2,verbose=2,minCount=1,epoch=25,loss='hs',autotuneValidationFile='valid.csv',autotuneMetric="f1:__label__baking")

#预测
val_pred = [model.predict(x)[0][0].split('__')[-1] for x in  train_df.iloc[-5000:]['text']]
print(f1_score(train_df['label'].values[-5000:].astype(str),val_pred,average='macro'))