Search code examples
pythonmachine-learningnlpnltktokenize

Unable to tokenize multiple columns in a dataframe


I have a table with both numeric and string data but in separate columns. The table is answers to a web form and contains empty cells. I want to use text processing on the string columns. I cannot drop the rows with empty cells so for the empty string columns, I replaced the NaN with aplhabet 'a'.

Sample data

colmun_name1    column_name2     column_name3 column_name4 classify
This is a cat   This is a dog    1            2            0
This is a rat   This is a mouse  45           32           1
a               Good mouse       0            0            0 

I used the following code to make sure all data in the string columns is actually string data.

df2=df[[column_name1, column_name2]]
for i in range(0,len(df2)):
cell=df2.iloc[i]
cell=str(str)
df2.iloc[i]=cell

Then when I tokenize, I get an error

    <ipython-input-64-24a99733ba19> in <module>
      1 from nltk.tokenize import word_tokenize
----> 2 tokenized_word=word_tokenize(df2)
      3 print(tokenized_word)

/anaconda3/lib/python3.6/site-packages/nltk/tokenize/__init__.py in word_tokenize(text, language, preserve_line)
    126     :type preserver_line: bool
    127     """
--> 128     sentences = [text] if preserve_line else sent_tokenize(text, language)
    129     return [token for sent in sentences
    130             for token in _treebank_word_tokenizer.tokenize(sent)]

/anaconda3/lib/python3.6/site-packages/nltk/tokenize/__init__.py in sent_tokenize(text, language)
     93     """
     94     tokenizer = load('tokenizers/punkt/{0}.pickle'.format(language))
---> 95     return tokenizer.tokenize(text)
     96 
     97 # Standard word tokenizer.

/anaconda3/lib/python3.6/site-packages/nltk/tokenize/punkt.py in tokenize(self, text, realign_boundaries)
   1239         Given a text, returns a list of the sentences in that text.
   1240         """
-> 1241         return list(self.sentences_from_text(text, realign_boundaries))
   1242 
   1243     def debug_decisions(self, text):

/anaconda3/lib/python3.6/site-packages/nltk/tokenize/punkt.py in sentences_from_text(self, text, realign_boundaries)
   1289         follows the period.
   1290         """
-> 1291         return [text[s:e] for s, e in self.span_tokenize(text, realign_boundaries)]
   1292 
   1293     def _slices_from_text(self, text):

/anaconda3/lib/python3.6/site-packages/nltk/tokenize/punkt.py in <listcomp>(.0)
   1289         follows the period.
   1290         """
-> 1291         return [text[s:e] for s, e in self.span_tokenize(text, realign_boundaries)]
   1292 
   1293     def _slices_from_text(self, text):

/anaconda3/lib/python3.6/site-packages/nltk/tokenize/punkt.py in span_tokenize(self, text, realign_boundaries)
   1279         if realign_boundaries:
   1280             slices = self._realign_boundaries(text, slices)
-> 1281         for sl in slices:
   1282             yield (sl.start, sl.stop)
   1283 

/anaconda3/lib/python3.6/site-packages/nltk/tokenize/punkt.py in _realign_boundaries(self, text, slices)
   1320         """
   1321         realign = 0
-> 1322         for sl1, sl2 in _pair_iter(slices):
   1323             sl1 = slice(sl1.start + realign, sl1.stop)
   1324             if not sl2:

/anaconda3/lib/python3.6/site-packages/nltk/tokenize/punkt.py in _pair_iter(it)
    311     """
    312     it = iter(it)
--> 313     prev = next(it)
    314     for el in it:
    315         yield (prev, el)

/anaconda3/lib/python3.6/site-packages/nltk/tokenize/punkt.py in _slices_from_text(self, text)
   1293     def _slices_from_text(self, text):
   1294         last_break = 0
-> 1295         for match in self._lang_vars.period_context_re().finditer(text):
   1296             context = match.group() + match.group('after_tok')
   1297             if self.text_contains_sentbreak(context):

TypeError: expected string or bytes-like object

I tried changing

df2=df[column_name1][column_name2]

But I get the same error.

What should I do?


Solution

  • Please see How to apply NLTK word_tokenize library on a Pandas dataframe for Twitter data?

    TL;DR

    # Creates a `colmun_name1_tokenized` column by 
    # taking the `colmun_name1` column and 
    # applying the word_tokenize function on every cell in the column. 
    
    >>> df['colmun_name1_tokenized'] = df['colmun_name1'].apply(word_tokenize)
    
    >>> df.head()
        colmun_name1     column_name2  column_name3  column_name4  classify  \
    0  This is a cat    This is a dog             1             2         0   
    1  This is a rat  This is a mouse            45            32         1   
    2              a       Good mouse             0             0         0   
    
      colmun_name1_tokenized  
    0     [This, is, a, cat]  
    1     [This, is, a, rat]  
    2                    [a]  
    

    If you need more than one column to be tokenized and you want to overwrite the column with the tokenized output:

    >>> with StringIO(file_str) as fin:
    ...     df = pd.read_csv(fin, sep='\t')
    ... 
    >>> for col_name in ['colmun_name1', 'column_name2']:
    ...     df[col_name] = df[col_name].apply(word_tokenize)
    ... 
    >>> df.head()
             colmun_name1          column_name2  column_name3  column_name4  \
    0  [This, is, a, cat]    [This, is, a, dog]             1             2   
    1  [This, is, a, rat]  [This, is, a, mouse]            45            32   
    2                 [a]         [Good, mouse]             0             0   
    
       classify  
    0         0  
    1         1  
    2         0  
    

    Just the code:

    from io import StringIO
    
    import pandas as pd
    
    from nltk import word_tokenize
    
    file_str = """colmun_name1\tcolumn_name2\tcolumn_name3\tcolumn_name4\tclassify
    This is a cat\tThis is a dog\t1\t2\t0
    This is a rat\tThis is a mouse\t45\t32\t1
    a\tGood mouse\t0\t0\t0 """
    
    with StringIO(file_str) as fin:
        df = pd.read_csv(fin, sep='\t')
    
    for col_name in ['colmun_name1', 'column_name2']:
        df[col_name] = df[col_name].apply(word_tokenize)