char_wb 的 Tf-idf 会忽略自定义预处理器？

Question

我有

import nltk
from nltk.stem.snowball import GermanStemmer

def my_tokenizer(doc):
   stemmer= GermanStemmer()

   return([stemmer.stem(t.lower()) for t in nltk.word_tokenize(doc) if 
   t.lower() not in my_stop_words])

text="hallo df sdfd"
singleTFIDF = TfidfVectorizer(analyzer='char_wb', ngram_range= 
(4,6),preprocessor=my_tokenizer, max_features=50).fit([str(text)])

从文档中可以清楚地看出，自定义 toenizer 仅适用于 analyzer=word。

我明白了

Traceback (most recent call last):
  File "TfidF.py", line 95, in <module>
    singleTFIDF = TfidfVectorizer(analyzer='char_wb', ngram_range=(4,6),preprocessor=my_tokenizer, max_features=50).fit([str(text)])
  File "C:\Users\chris1\Anaconda3\envs\master\lib\site-packages\sklearn\feature_extraction\text.py", line 185, in _char_wb_ngrams
    text_document = self._white_spaces.sub(" ", text_document)
TypeError: expected string or bytes-like object

Answer 1

你必须加入单词，然后 return 一个字符串。试试这个！

return(' '.join ([stemmer.stem(t.lower()) for t in nltk.word_tokenize(doc) if 
   t.lower() not in my_stop_words]))

char_wb 的 Tf-idf 会忽略自定义预处理器？

Tf-idf with char_wb ignores custom preporcessor?

python

scikit-learn

tfidfvectorizer