Preparation¶

In [1]:

import sys
!{sys.executable} -m pip install --user pandas numpy sklearn nltk

Requirement already satisfied: pandas in /usr/lib64/python3.8/site-packages (0.25.3)
Requirement already satisfied: numpy in /usr/lib64/python3.8/site-packages (1.18.4)
Requirement already satisfied: sklearn in /home/pasha/.local/lib/python3.8/site-packages (0.0)
Requirement already satisfied: nltk in /home/pasha/.local/lib/python3.8/site-packages (3.4.5)
Requirement already satisfied: python-dateutil>=2.6.1 in /usr/lib/python3.8/site-packages (from pandas) (2.8.0)
Requirement already satisfied: pytz>=2017.2 in /usr/lib/python3.8/site-packages (from pandas) (2020.1)
Requirement already satisfied: scikit-learn in /home/pasha/.local/lib/python3.8/site-packages (from sklearn) (0.22.2.post1)
Requirement already satisfied: six in /usr/lib/python3.8/site-packages (from nltk) (1.14.0)
Requirement already satisfied: joblib>=0.11 in /home/pasha/.local/lib/python3.8/site-packages (from scikit-learn->sklearn) (0.14.1)
Requirement already satisfied: scipy>=0.17.0 in /usr/lib64/python3.8/site-packages (from scikit-learn->sklearn) (1.4.1)

In [2]:

import pandas as pd
import numpy as np
from sklearn.feature_extraction.text import TfidfVectorizer

In [3]:

# Define constants
tfidf = TfidfVectorizer(stop_words='english')

In [4]:

# By https://ru.stackoverflow.com/questions/995616/Как-сделать-tf-idf-для-русских-текстов

#import nltk
#from nltk.corpus import stopwords as nltk_stopwords

#nltk.download('stopwords')
#stopwords = set(nltk_stopwords.words('russian') )
#tfidf = TfidfVectorizer(stop_words=stopwords)

In [5]:

# Minimal example:
articles = pd.read_csv('data/TF-IDF/TF-IDF-min.csv.Ru', index_col='№')
# articles = pd.read_csv('data/TF-IDF/TF-IDF-min.csv.En', index_col='№')
# Full real file:
# articles = pd.read_csv('RDC-135_articles_golden_set_mapping.csv', index_col='№')
articles

Out[5]:

	Article
№
1	Раз, два, три, четыре
2	вышел зайчик погулять
3	вдруг охотник выбегает
4	прямо в зайчик стреляет
5	пуляет прямо в зайчик

In [6]:

tfidf_matrix = tfidf.fit_transform(articles['Article'])
tfidf_matrix.toarray() # or res.todense()

Out[6]:

array([[0.        , 0.        , 0.        , 0.5       , 0.        ,
        0.        , 0.        , 0.        , 0.        , 0.5       ,
        0.        , 0.5       , 0.5       ],
       [0.        , 0.        , 0.63907044, 0.        , 0.42799292,
        0.        , 0.63907044, 0.        , 0.        , 0.        ,
        0.        , 0.        , 0.        ],
       [0.57735027, 0.57735027, 0.        , 0.        , 0.        ,
        0.57735027, 0.        , 0.        , 0.        , 0.        ,
        0.        , 0.        , 0.        ],
       [0.        , 0.        , 0.        , 0.        , 0.4622077 ,
        0.        , 0.        , 0.55681615, 0.        , 0.        ,
        0.69015927, 0.        , 0.        ],
       [0.        , 0.        , 0.        , 0.        , 0.4622077 ,
        0.        , 0.        , 0.55681615, 0.69015927, 0.        ,
        0.        , 0.        , 0.        ]])

In [7]:

tfidf_matrix.shape

Out[7]:

(5, 13)

In [8]:

tfidf.stop_words

Out[8]:

'english'

Вычисление¶

Самые популярные слова¶

In [9]:

# By https://www.rupython.com/sklearn-tfidf-transformer-tf-idf-33655.html
# Вышеупомянутый X имеет значения TF-IDF всех документов в корпусе. Это большая разреженная матрица.
# Теперь,
tfidf.get_feature_names()
# это дает вам список всех токенов или n-граммов или слов. Для первого документа в вашем корпусе, 

Out[9]:

['вдруг',
 'выбегает',
 'вышел',
 'два',
 'зайчик',
 'охотник',
 'погулять',
 'прямо',
 'пуляет',
 'раз',
 'стреляет',
 'три',
 'четыре']

In [10]:

# By https://stackoverflow.com/questions/37593293/what-is-the-simplest-way-to-get-tfidf-with-pandas-dataframe#comment72191707_37593408
# v.get_feature_names() will give you the list of feature names.
# v.vocabulary_ will give you a dict with feature names as keys and their index in the matrix produced as values.
tfidf.vocabulary_

Out[10]:

{'раз': 9,
 'два': 3,
 'три': 11,
 'четыре': 12,
 'вышел': 2,
 'зайчик': 4,
 'погулять': 6,
 'вдруг': 0,
 'охотник': 5,
 'выбегает': 1,
 'прямо': 7,
 'стреляет': 10,
 'пуляет': 8}

In [11]:

# By https://www.rupython.com/sklearn-tfidf-transformer-tf-idf-33655.html
# это дает вам список всех токенов или n-граммов или слов. Для первого документа в вашем корпусе, 
# Позволяет распечатать их:
def print_word_ratings_by_document(doc=0):
    feature_names = tfidf.get_feature_names()
    feature_index = tfidf_matrix[doc,:].nonzero()[1]
    tfidf_scores = zip(feature_index, [tfidf_matrix[doc, x] for x in feature_index]) 
    for w, s in [(feature_names[i], s) for (i, s) in tfidf_scores]:
        print (w, s)

for i, _ in articles.iterrows():
    print("### doc {}: ###".format(i-1))
    print_word_ratings_by_document(i-1)

### doc 0: ###
четыре 0.5
три 0.5
два 0.5
раз 0.5
### doc 1: ###
погулять 0.6390704413963749
зайчик 0.42799292268317357
вышел 0.6390704413963749
### doc 2: ###
выбегает 0.5773502691896258
охотник 0.5773502691896258
вдруг 0.5773502691896258
### doc 3: ###
стреляет 0.6901592662889633
прямо 0.5568161504458247
зайчик 0.46220770413113277
### doc 4: ###
пуляет 0.6901592662889633
прямо 0.5568161504458247
зайчик 0.46220770413113277

In [12]:

# https://ru.stackoverflow.com/questions/772859/tfidfvectorizer/773018#773018
# Топ 10 самых популярных слов:
N=10
idx = np.ravel(tfidf_matrix.sum(axis=0).argsort(axis=1))[::-1][:N]
top_10_words = np.array(tfidf.get_feature_names())[idx].tolist()
top_10_words

Out[12]:

['зайчик',
 'прямо',
 'стреляет',
 'пуляет',
 'погулять',
 'вышел',
 'охотник',
 'выбегает',
 'вдруг',
 'четыре']

Разбор¶

In [13]:

tfidf_matrix.shape

Out[13]:

(5, 13)

In [14]:

tfidf_matrix.todense()

Out[14]:

matrix([[0.        , 0.        , 0.        , 0.5       , 0.        ,
         0.        , 0.        , 0.        , 0.        , 0.5       ,
         0.        , 0.5       , 0.5       ],
        [0.        , 0.        , 0.63907044, 0.        , 0.42799292,
         0.        , 0.63907044, 0.        , 0.        , 0.        ,
         0.        , 0.        , 0.        ],
        [0.57735027, 0.57735027, 0.        , 0.        , 0.        ,
         0.57735027, 0.        , 0.        , 0.        , 0.        ,
         0.        , 0.        , 0.        ],
        [0.        , 0.        , 0.        , 0.        , 0.4622077 ,
         0.        , 0.        , 0.55681615, 0.        , 0.        ,
         0.69015927, 0.        , 0.        ],
        [0.        , 0.        , 0.        , 0.        , 0.4622077 ,
         0.        , 0.        , 0.55681615, 0.69015927, 0.        ,
         0.        , 0.        , 0.        ]])

In [15]:

# Doc: https://docs.scipy.org/doc/numpy/reference/generated/numpy.sum.html#numpy.sum
tfidf_matrix.sum(axis=0)  # Сумма по столбцам

Out[15]:

matrix([[0.57735027, 0.57735027, 0.63907044, 0.5       , 1.35240833,
         0.57735027, 0.63907044, 1.1136323 , 0.69015927, 0.5       ,
         0.69015927, 0.5       , 0.5       ]])

In [16]:

# Doc: https://docs.scipy.org/doc/numpy/reference/generated/numpy.argsort.html
tfidf_matrix.sum(axis=0).argsort(axis=1)  # Возвращает *индексы*, по возрастанию значений элементов

Out[16]:

matrix([[ 3,  9, 11, 12,  0,  1,  5,  2,  6,  8, 10,  7,  4]])

In [17]:

# Doc: https://docs.scipy.org/doc/numpy/reference/generated/numpy.ravel.html
np.ravel(tfidf_matrix.sum(axis=0).argsort(axis=1))

Out[17]:

array([ 3,  9, 11, 12,  0,  1,  5,  2,  6,  8, 10,  7,  4])

In [18]:

np.ravel(tfidf_matrix.sum(axis=0).argsort(axis=1))[::-1]  # Reverse list (DESC)

Out[18]:

array([ 4,  7, 10,  8,  6,  2,  5,  1,  0, 12, 11,  9,  3])

In [19]:

np.ravel(tfidf_matrix.sum(axis=0).argsort(axis=1))[::-1][:N] # Take top N elements

Out[19]:

array([ 4,  7, 10,  8,  6,  2,  5,  1,  0, 12])

In [20]:

np.ravel(tfidf_matrix.sum(axis=0).argsort(axis=1))[-N:][::-1]

Out[20]:

array([ 4,  7, 10,  8,  6,  2,  5,  1,  0, 12])

Поиск самого похожего документа¶

In [21]:

similarities = (tfidf_matrix * tfidf_matrix.T).A[-1,:-1]
similarities

Out[21]:

array([0.        , 0.19782163, 0.        , 0.52368019])

In [22]:

# Doc: https://docs.scipy.org/doc/numpy/reference/generated/numpy.argmax.html
#  numpy.argmax(a, axis=None, out=None)
# Returns the indices of the maximum values along an axis.
max_sim_position = np.argmax(similarities)
max_sim_position

Out[22]:

In [23]:

max_sim = max(similarities)
max_sim

Out[23]:

0.52368018715548

In [24]:

similarities[max_sim_position]

Out[24]:

0.52368018715548

Разбор¶

In [25]:

tfidf_matrix.todense()

Out[25]:

matrix([[0.        , 0.        , 0.        , 0.5       , 0.        ,
         0.        , 0.        , 0.        , 0.        , 0.5       ,
         0.        , 0.5       , 0.5       ],
        [0.        , 0.        , 0.63907044, 0.        , 0.42799292,
         0.        , 0.63907044, 0.        , 0.        , 0.        ,
         0.        , 0.        , 0.        ],
        [0.57735027, 0.57735027, 0.        , 0.        , 0.        ,
         0.57735027, 0.        , 0.        , 0.        , 0.        ,
         0.        , 0.        , 0.        ],
        [0.        , 0.        , 0.        , 0.        , 0.4622077 ,
         0.        , 0.        , 0.55681615, 0.        , 0.        ,
         0.69015927, 0.        , 0.        ],
        [0.        , 0.        , 0.        , 0.        , 0.4622077 ,
         0.        , 0.        , 0.55681615, 0.69015927, 0.        ,
         0.        , 0.        , 0.        ]])

In [26]:

tfidf_matrix.shape

Out[26]:

(5, 13)

In [27]:

type(tfidf_matrix)

Out[27]:

scipy.sparse.csr.csr_matrix

In [52]:

mul = tfidf_matrix * tfidf_matrix.T
mul.todense()

Out[52]:

matrix([[1.        , 0.        , 0.        , 0.        , 0.        ],
        [0.        , 1.        , 0.        , 0.19782163, 0.19782163],
        [0.        , 0.        , 1.        , 0.        , 0.        ],
        [0.        , 0.19782163, 0.        , 1.        , 0.52368019],
        [0.        , 0.19782163, 0.        , 0.52368019, 1.        ]])

In [53]:

mul.todense().shape

Out[53]:

(5, 5)

In [54]:

scores = mul.A
scores

Out[54]:

array([[1.        , 0.        , 0.        , 0.        , 0.        ],
       [0.        , 1.        , 0.        , 0.19782163, 0.19782163],
       [0.        , 0.        , 1.        , 0.        , 0.        ],
       [0.        , 0.19782163, 0.        , 1.        , 0.52368019],
       [0.        , 0.19782163, 0.        , 0.52368019, 1.        ]])

In [32]:

# We will search maximum, so do not willing match to himself:
np.fill_diagonal(scores, -1)
scores

Out[32]:

array([[-1.        ,  0.        ,  0.        ,  0.        ,  0.        ],
       [ 0.        , -1.        ,  0.        ,  0.19782163,  0.19782163],
       [ 0.        ,  0.        , -1.        ,  0.        ,  0.        ],
       [ 0.        ,  0.19782163,  0.        , -1.        ,  0.52368019],
       [ 0.        ,  0.19782163,  0.        ,  0.52368019, -1.        ]])

In [33]:

type(scores)

Out[33]:

numpy.ndarray

In [34]:

scores_df = pd.DataFrame(scores, index=articles.index)
scores_df

Out[34]:

	0	1	2	3	4
№
1	-1.0	0.000000	0.0	0.000000	0.000000
2	0.0	-1.000000	0.0	0.197822	0.197822
3	0.0	0.000000	-1.0	0.000000	0.000000
4	0.0	0.197822	0.0	-1.000000	0.523680
5	0.0	0.197822	0.0	0.523680	-1.000000

In [59]:

# By https://gist.github.com/RZachLamberty/1ed47cd0e2d0d968f7cdbd3d53a50f4c
# you can calculate cosine similarity easily given this
(tfidf_matrix @ tfidf_matrix.T).todense()

Out[59]:

matrix([[1.        , 0.        , 0.        , 0.        , 0.        ],
        [0.        , 1.        , 0.        , 0.19782163, 0.19782163],
        [0.        , 0.        , 1.        , 0.        , 0.        ],
        [0.        , 0.19782163, 0.        , 1.        , 0.52368019],
        [0.        , 0.19782163, 0.        , 0.52368019, 1.        ]])

In [56]:

# Doc: https://docs.scipy.org/doc/numpy/reference/generated/numpy.matrix.A.html
# Return self as an ndarray object.
# Equivalent to np.asarray(self)
mul.A

Out[56]:

array([[1.        , 0.        , 0.        , 0.        , 0.        ],
       [0.        , 1.        , 0.        , 0.19782163, 0.19782163],
       [0.        , 0.        , 1.        , 0.        , 0.        ],
       [0.        , 0.19782163, 0.        , 1.        , 0.52368019],
       [0.        , 0.19782163, 0.        , 0.52368019, 1.        ]])

In [57]:

mul.A[-1,:-1] # Последняя строка

Out[57]:

array([0.        , 0.19782163, 0.        , 0.52368019])

In [38]:

mul.A[-1,0:] # Последняя строка

Out[38]:

array([0.        , 0.19782163, 0.        , 0.52368019, 1.        ])

In [39]:

mul.A[-1,:-1] # Последняя строка без последнего элемента.
# Матрица диагональная (остальные не имеют значения, зеркально повторяются), содержит веса всех со всеми.
# Последний элемент, выкидывается потому что это матч самого к себе, если мы рассматриваем максимальный дубль
# для последнего в наборе документа (так было в функции get_similarities из GOJI)

# Получается что реально нужно выкидывать по индексу того, для которого ищется сравнение!

Out[39]:

array([0.        , 0.19782163, 0.        , 0.52368019])

Обобщение варианта - без пересчёта матрицы каждый раз!¶

In [40]:

# Получается что реально нужно выкидывать по индексу того, для которого ищется сравнение!
forDocNo=1 # Expect match doc 3<>4

In [41]:

scores

Out[41]:

array([[-1.        ,  0.        ,  0.        ,  0.        ,  0.        ],
       [ 0.        , -1.        ,  0.        ,  0.19782163,  0.19782163],
       [ 0.        ,  0.        , -1.        ,  0.        ,  0.        ],
       [ 0.        ,  0.19782163,  0.        , -1.        ,  0.52368019],
       [ 0.        ,  0.19782163,  0.        ,  0.52368019, -1.        ]])

In [42]:

s = scores[forDocNo]
max_sim_position = np.argmax(s)

(max_sim_position, s[max_sim_position])

Out[42]:

(3, 0.19782162617776308)

Для каждого документа, в DataFrame add "Max TF/IDF DUP score" and "Max DUP score docId"¶

In [43]:

articles

# Should be matched:
# id № Dup№
#  0 1 -
#  1 2 (4 and 5)
#  2 3 -
#  3 4 5
#  4 5 4

Out[43]:

	Article
№
1	Раз, два, три, четыре
2	вышел зайчик погулять
3	вдруг охотник выбегает
4	прямо в зайчик стреляет
5	пуляет прямо в зайчик

In [44]:

scores

Out[44]:

array([[-1.        ,  0.        ,  0.        ,  0.        ,  0.        ],
       [ 0.        , -1.        ,  0.        ,  0.19782163,  0.19782163],
       [ 0.        ,  0.        , -1.        ,  0.        ,  0.        ],
       [ 0.        ,  0.19782163,  0.        , -1.        ,  0.52368019],
       [ 0.        ,  0.19782163,  0.        ,  0.52368019, -1.        ]])

In [45]:

scores_df

Out[45]:

	0	1	2	3	4
№
1	-1.0	0.000000	0.0	0.000000	0.000000
2	0.0	-1.000000	0.0	0.197822	0.197822
3	0.0	0.000000	-1.0	0.000000	0.000000
4	0.0	0.197822	0.0	-1.000000	0.523680
5	0.0	0.197822	0.0	0.523680	-1.000000

In [46]:

scores_df.loc[3]

Out[46]:

0    0.0
1    0.0
2   -1.0
3    0.0
4    0.0
Name: 3, dtype: float64

In [47]:

scores_df.iloc[3]

Out[47]:

0    0.000000
1    0.197822
2    0.000000
3   -1.000000
4    0.523680
Name: 4, dtype: float64

In [48]:

# By https://stackoverflow.com/questions/26658240/getting-the-index-of-a-row-in-a-pandas-apply-function/48819898#48819898
# index available as row.name
def most_similar(row):
#     print('====')
#     print('row.name={}; scores_df.loc[row.name].idxmax()={}; scores_df.iloc[scores_df.loc[row.name].idxmax()].name={}'.format(
#         row.name
#         ,scores_df.loc[row.name].idxmax()
#         ,scores_df.iloc[scores_df.loc[row.name].idxmax()].name
#         )
#     )
#     print('row.loc={}; row.iloc={}'.format(row.loc, row.iloc))
    max_similar_doc = scores_df.iloc[scores_df.loc[row.name].idxmax()].name # Array index (.iloc) into DataFrame index (.name)
    max_similar_score = scores_df.loc[row.name].max()
    return ((max_similar_doc if max_similar_score > 0 else -1), max_similar_score)

# articles['Max DUP score docId'] = articles.apply(lambda i: np.argmax(scores), axis=1)
# articles['Max DUP score docId'] = articles.apply(lambda i: type(i.index), axis=1)
articles[['Max DUP score docId', 'Max TF/IDF DUP score']] = articles.apply(most_similar, axis=1, result_type='expand')
articles['Max DUP score docId'] = articles['Max DUP score docId'].astype('int32')
articles

Out[48]:

	Article	Max DUP score docId	Max TF/IDF DUP score
№
1	Раз, два, три, четыре	-1	0.000000
2	вышел зайчик погулять	4	0.197822
3	вдруг охотник выбегает	-1	0.000000
4	прямо в зайчик стреляет	5	0.523680
5	пуляет прямо в зайчик	4	0.523680

In [49]:

( scores_df[3], scores_df.loc[3] )

Out[49]:

(№
 1    0.000000
 2    0.197822
 3    0.000000
 4   -1.000000
 5    0.523680
 Name: 3, dtype: float64,
 0    0.0
 1    0.0
 2   -1.0
 3    0.0
 4    0.0
 Name: 3, dtype: float64)

Python constructions¶

Apply with 2 columns at once¶

By https://stackoverflow.com/questions/16236684/apply-pandas-function-to-column-to-create-multiple-new-columns/52363890#52363890 +By https://stackoverflow.com/questions/16236684/apply-pandas-function-to-column-to-create-multiple-new-columns/52363890#comment106834440_16242202 for column naming

In [50]:

# articles[['a', 'b']] = articles.apply(lambda i: [1, 2], axis=1, result_type='expand')
# articles

In [51]:

def dump_xml_file(row):
    print(type(row))
    print(row['File name'])
    with open('articles/' + row['File name'].replace('.xml', '.txt'), 'w+') as file:
        file.write(row['Article'])

articles.apply(dump_xml_file, axis=1)

<class 'pandas.core.series.Series'>
<class 'pandas.core.series.Series'>

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/usr/lib64/python3.8/site-packages/pandas/core/indexes/base.py in get_value(self, series, key)
   4735             try:
-> 4736                 return libindex.get_value_box(s, key)
   4737             except IndexError:

pandas/_libs/index.pyx in pandas._libs.index.get_value_box()

pandas/_libs/index.pyx in pandas._libs.index.get_value_at()

pandas/_libs/util.pxd in pandas._libs.util.get_value_at()

pandas/_libs/util.pxd in pandas._libs.util.validate_indexer()

TypeError: 'str' object cannot be interpreted as an integer

During handling of the above exception, another exception occurred:

KeyError                                  Traceback (most recent call last)
<ipython-input-51-38f66aea0c91> in <module>
      5         file.write(row['Article'])
      6 
----> 7 articles.apply(dump_xml_file, axis=1)

/usr/lib64/python3.8/site-packages/pandas/core/frame.py in apply(self, func, axis, broadcast, raw, reduce, result_type, args, **kwds)
   6926             kwds=kwds,
   6927         )
-> 6928         return op.get_result()
   6929 
   6930     def applymap(self, func):

/usr/lib64/python3.8/site-packages/pandas/core/apply.py in get_result(self)
    184             return self.apply_raw()
    185 
--> 186         return self.apply_standard()
    187 
    188     def apply_empty_result(self):

/usr/lib64/python3.8/site-packages/pandas/core/apply.py in apply_standard(self)
    290 
    291         # compute the result using the series generator
--> 292         self.apply_series_generator()
    293 
    294         # wrap results

/usr/lib64/python3.8/site-packages/pandas/core/apply.py in apply_series_generator(self)
    319             try:
    320                 for i, v in enumerate(series_gen):
--> 321                     results[i] = self.f(v)
    322                     keys.append(v.name)
    323             except Exception as e:

<ipython-input-51-38f66aea0c91> in dump_xml_file(row)
      1 def dump_xml_file(row):
      2     print(type(row))
----> 3     print(row['File name'])
      4     with open('articles/' + row['File name'].replace('.xml', '.txt'), 'w+') as file:
      5         file.write(row['Article'])

/usr/lib64/python3.8/site-packages/pandas/core/series.py in __getitem__(self, key)
   1069         key = com.apply_if_callable(key, self)
   1070         try:
-> 1071             result = self.index.get_value(self, key)
   1072 
   1073             if not is_scalar(result):

/usr/lib64/python3.8/site-packages/pandas/core/indexes/base.py in get_value(self, series, key)
   4742                     raise InvalidIndexError(key)
   4743                 else:
-> 4744                     raise e1
   4745             except Exception:  # pragma: no cover
   4746                 raise e1

/usr/lib64/python3.8/site-packages/pandas/core/indexes/base.py in get_value(self, series, key)
   4728         k = self._convert_scalar_indexer(k, kind="getitem")
   4729         try:
-> 4730             return self._engine.get_value(s, k, tz=getattr(series.dtype, "tz", None))
   4731         except KeyError as e1:
   4732             if len(self) > 0 and (self.holds_integer() or self.is_boolean()):

pandas/_libs/index.pyx in pandas._libs.index.IndexEngine.get_value()

pandas/_libs/index.pyx in pandas._libs.index.IndexEngine.get_value()

pandas/_libs/index.pyx in pandas._libs.index.IndexEngine.get_loc()

pandas/_libs/hashtable_class_helper.pxi in pandas._libs.hashtable.PyObjectHashTable.get_item()

pandas/_libs/hashtable_class_helper.pxi in pandas._libs.hashtable.PyObjectHashTable.get_item()

KeyError: ('File name', 'occurred at index 1')

Preparation¶

Вычисление¶

Самые популярные слова¶

Разбор¶

Поиск самого похожего документа¶

Разбор¶

Обобщение варианта - без пересчёта матрицы каждый раз!¶

Для каждого документа, в DataFrame add "Max TF/IDF DUP score" and "Max DUP score docId"¶

Python constructions¶

Apply with 2 columns at once¶

Docs and links¶