Python / Pandas / spacy - iterate over a DataFrame and count the number of pos_ tags

问题

i have a Pandas Dataframe with some Texts from an Author and want to do some statistical stuff with the sum of the different word types.

Dataframe - my data :

>>> data
             name                   style                                              text     year       year_dt
number  
0001    Demetrius                       D   Demetrius an der russischen Grenze Er ist vo...     1805    1805-01-01
0002    Der versöhnte Menschenfeind     D   Der versöhnte Menschenfeind -Fragment Gegend...     1790    1790-01-01
0003    Die Braut von Messina           D   Die Braut von Messina oder die feindlichen B...     1803    1803-01-01

Some months ago i wrote a function that iterates from line to line of the df, take the name and the content of "the book", made a pos tagging from spacy, and count the number of nouns, adjectives and verbs for the beginning. after that, the number is stored in a new column.

my function:

import spacy
from spacy.lang.de import German
from collections import defaultdict
nlp = spacy.load('de')

def calculate_the_word_types(data):
    nouns = defaultdict(lambda: 0)
    verbs = defaultdict(lambda: 0)
    adjectives = defaultdict(lambda: 0)

    # count all tokens, but not the punctuations
    for i, row in data.iterrows():
        doc = nlp(row["name"] + " " + row["text"])
    data.set_value(i, "nr_token", len(list(map(lambda x: x.text, 
                                     filter(lambda x: x.pos_ != 'PUNCT', doc)))))

    # count only the adjectives
    for a in map(lambda x: x.lemma_, filter(lambda x: x.pos_ == 'ADJ', doc)):
        adjectives[a] += 1
    data.set_value(i, "nr_adj", len(list(map(lambda x: x.text, 
                                     filter(lambda x: x.pos_ == 'ADJ', doc)))))  

    # count only the nouns
    for n in map(lambda x: x.lemma_, filter(lambda x: x.pos_ == 'NOUN', doc)):
        nouns[n] +=1
    data.set_value(i, "nr_noun", len(list(map(lambda x: x.text, 
                                     filter(lambda x: x.pos_ == 'NOUN', doc)))))

    # count only the verbs
    for v in map(lambda x: x.lemma_, filter(lambda x: (x.pos_ == 'AUX') | (x.pos_ == 'VERB'), doc)):
        verbs[v] += 1
    data.set_value(i, "nr_verb", len(list(map(lambda x: x.text, 
                                     filter(lambda x: (x.pos_ == 'AUX') | (x.pos_ == 'VERB'), doc)))))  

    return data

the output

>>> data
           name style      text     year       year_dt  nr_token  br_adj   nr_noun   nr_verb
number  
0001    Deme...     D   Deme...     1805    1805-01-01       NaN     NaN       NaN       NaN
0002    Der ...     D   Der ...     1790    1790-01-01       NaN     NaN       NaN       NaN
0003    Die ...     D   Die ...     1803    1803-01-01    7127.0   584.0    1328.0    1286.0

i think this worked back then, but not now. because my function output is the following and through testing i know, that it works, but the numbers are always in the last line only, so it overwrites itself, i think.

where is the failiure? every tipps or suggestions to improve this function or maybe some easier solution are welcome!

回答1:

Indent your setter so that is it inside the outer for loop.

# count all tokens, but not the punctuations
for i, row in data.iterrows():
    doc = nlp(row["name"] + " " + row["text"])
    data.set_value(i, "nr_token", len(list(map(lambda x: x.text, 
                                 filter(lambda x: x.pos_ != 'PUNCT', doc)))))

    # count only the adjectives
    for a in map(lambda x: x.lemma_, filter(lambda x: x.pos_ == 'ADJ', doc)):
        adjectives[a] += 1
    data.set_value(i, "nr_adj", len(list(map(lambda x: x.text, 
                                 filter(lambda x: x.pos_ == 'ADJ', doc)))))  

    # count only the nouns
    for n in map(lambda x: x.lemma_, filter(lambda x: x.pos_ == 'NOUN', doc)):
        nouns[n] +=1
    data.set_value(i, "nr_noun", len(list(map(lambda x: x.text, 
                                 filter(lambda x: x.pos_ == 'NOUN', doc)))))

    # count only the verbs
    for v in map(lambda x: x.lemma_, filter(lambda x: (x.pos_ == 'AUX') | (x.pos_ == 'VERB'), doc)):
        verbs[v] += 1
    data.set_value(i, "nr_verb", len(list(map(lambda x: x.text, 
                                 filter(lambda x: (x.pos_ == 'AUX') | (x.pos_ == 'VERB'), doc)))))

来源：https://stackoverflow.com/questions/54390101/python-pandas-spacy-iterate-over-a-dataframe-and-count-the-number-of-pos