Coverage for dariah_topics/preprocessing.py : 92%

#!/usr/bin/env python3 

# -*- coding: utf-8 -*- 

""" 

Processing Text Data, Creating Matrices and Cleaning Corpora 

============================================================ 

Functions of this module are for **preprocessing purpose**. You can read text \ 

files, `tokenize <https://en.wikipedia.org/wiki/Tokenization_(lexical_analysis)>`_ \ 

and segment documents, create `document-term matrices <https://en.wikipedia.org/wiki/Document-term_matrix>`_, \ 

determine and remove features and read existing matrices. Recurrent variable names are \ 

based on the following conventions: 

1. Corpora: 

*********** 

* ``corpus`` means an iterable containing at least one ``document`` or ``dkpro_document``. 

* ``document`` means one single string containing all characters of a text \ 

file, including whitespaces, punctuations, etc. 

* ``dkpro_document`` means a pandas DataFrame containing tokens and additional \ 

information, e.g. *part-of-speech tags* or *lemmas*. 

* ``tokenized_corpus`` means an iterable containing at least one ``tokenized_document``. 

* ``tokenized_document`` means an iterable containing tokens of a ``document``. 

* ``clean_tokenized_corpus`` means an iterable containing at least one ``clean_tokenized_document``. 

* ``clean_tokenized_document`` means an iterable containing only specific \ 

tokens (e.g. no *stopwords* or hapax *legomena*) of a ``tokenized_document``.  

* ``document_labels`` means an iterable containing names of each ``document`` \ 

and must have as much elements as ``corpus``, ``tokenized_corpus`` or \ 

``clean_tokenized_corpus``, respectively. 

Furthermore, if a document is chunked into smaller segments, each segment counts 

as one document. 

2. Data models: 

*************** 

* ``document_term_matrix`` means either a pandas DataFrame with rows corresponding to \ 

``document_labels`` and columns to types (distinct tokens in the corpus). The \ 

single values are token frequencies, or a pandas DataFrame with a MultiIndex \ 

and only one column corresponding to word frequencies. The first column of the \ 

MultiIndex corresponds to a document ID (based on ``document_labels``) and the \ 

second column to a type ID. 

Contents: 

********* 

* :func:`create_document_term_matrix()` creates a document-term matrix, for either \ 

large or small corpora. 

* :func:`duplicate_document_label()` duplicates a ``document_label`` with consecutive \ 

numbers. 

* :func:`filter_dkpro_document()` filters a ``dkpro_document`` by specific \ 

*part-of-speech tags*. 

* :func:`find_hapax_legomena()` determines *hapax legomena* based on frequencies \ 

of a ``document_term_matrix``. 

* :func:`find_stopwords()` determines *most frequent words* based on frequencies \ 

of a ``document_term_matrix``. 

* :func:`read_from_pathlist()` reads one or multiple files based on a pathlist. 

* :func:`segment()` is a wrapper for :func:`segment_fuzzy()` and segments a \ 

``tokenized_document`` into segments of a certain number of tokens, respecting existing chunks. 

* :func:`segment_fuzzy()` segments a ``tokenized_document``, tolerating existing \ 

chunks (like paragraphs). 

* :func:`split_paragraphs()` splits a ``document`` by paragraphs. 

* :func:`tokenize()` tokenizes a ``document`` based on a Unicode regular expression. 

* :func:`remove_features()` removes features from a ``document_term_matrix``. 

""" 

from collections import Counter, defaultdict 

import csv 

from itertools import chain 

import logging 

from lxml import etree 

import numpy as np 

import os 

import pandas as pd 

import regex 

log = logging.getLogger('preprocessing') 

log.addHandler(logging.NullHandler()) 

logging.basicConfig(level=logging.ERROR, 

format='%(levelname)s %(name)s: %(message)s') 

regular_expression = r'\p{Letter}+\p{Punctuation}?\p{Letter}+' 

def read_from_txt(doclist): 

"""Opens TXT files using file paths. 

Description: 

With this function you can read plain text files. Commit a list of 

full paths or one single path as argument. 

Use the function `create_document_list()` to create a list of your text 

files. 

Args: 

doclist Union(list[str], str): List of all documents in the corpus 

or single path to TXT file. 

Yields: 

Document. 

Todo: 

* Separate metadata (author, header) 

Example: 

>>> list(read_from_txt('corpus_txt/Doyle_AScandalinBohemia.txt'))[0][:20] 

'A SCANDAL IN BOHEMIA' 

""" 

log.info("Accessing TXT documents ...") 

if isinstance(doclist, str): 

with open(doclist, 'r', encoding='utf-8') as f: 

yield f.read() 

elif isinstance(doclist, list): 

for file in doclist: 

with open(file, 'r', encoding='utf-8') as f: 

yield f.read() 

def read_from_tei(doclist): 

"""Opens TEI XML files using file paths. 

Description: 

With this function you can read TEI encoded XML files. Commit a list of 

full paths or one single path as argument. 

Use the function `create_document_list()` to create a list of your XML 

files. 

Args: 

doclist Union(list[str], str): List of all documents in the corpus 

or single path to TEI XML file. 

Yields: 

Document. 

Todo: 

* Seperate metadata (author, header)? 

Example: 

>>> list(read_from_tei('corpus_tei/Schnitzler_Amerika.xml'))[0][142:159] 

'Arthur Schnitzler' 

""" 

log.info("Accessing TEI XML documents ...") 

if not isinstance(doclist, list): 

doclist = [doclist] 

ns = dict(tei='http://www.tei-c.org/ns/1.0') 

for file in doclist: 

tree = etree.parse(file) 

text_el = tree.xpath('//tei:text', namespaces=ns)[0] 

yield "".join(text_el.xpath('.//text()')) 

def read_from_csv(doclist, columns=['ParagraphId', 'TokenId', 'Lemma', 'CPOS', 'NamedEntity']): 

"""Opens CSV files using file paths. 

Description: 

With this function you can read CSV files generated by `DARIAH-DKPro-Wrapper`_, 

a tool for natural language processing. Commit a list of full paths or 

one single path as argument. You also have the ability to select certain 

columns. 

Use the function `create_document_list()` to create a list of your CSV 

files. 

.. _DARIAH-DKPro-Wrapper: 

https://github.com/DARIAH-DE/DARIAH-DKPro-Wrapper 

Args: 

doclist Union(list[str], str): List of all documents in the corpus 

or single path to CSV file. 

columns (list[str]): List of CSV column names. 

Defaults to '['ParagraphId', 'TokenId', 'Lemma', 'CPOS', 'NamedEntity']'. 

Yields: 

Document. 

Todo: 

* Seperate metadata (author, header)? 

Example: 

>>> list(read_from_csv('corpus_csv/Doyle_AScandalinBohemia.txt.csv'))[0][:4] # doctest: +NORMALIZE_WHITESPACE 

ParagraphId TokenId Lemma CPOS NamedEntity 

0 0 0 a ART _ 

1 0 1 scandal NP _ 

2 0 2 in PP _ 

3 0 3 bohemia NP _ 

""" 

log.info("Accessing CSV documents ...") 

if isinstance(doclist, str): 

doclist = [doclist] 

for file in doclist: 

df = pd.read_csv(file, sep='\t', quoting=csv.QUOTE_NONE) 

yield df[columns] 

def tokenize(doc_txt, expression=regular_expression, lower=True, simple=False): 

"""Tokenizes with Unicode Regular Expressions. 

Description: 

With this function you can tokenize a document with a regular expression. 

You also have the ability to commit your own regular expression. The default 

expression is '\p{Letter}+\p{Punctuation}?\p{Letter}+', which means one or 

more letters, followed by one or no punctuation, followed by one or more 

letters. So one letter words won't match. 

In case you want to lower alls tokens, set the argument `lower` to True (it 

is by default). 

If you want a very simple and primitive tokenization, set the argument 

`simple` to True. 

Use the functions `read_from_txt()`, `read_from_tei()` or `read_from_csv()` 

to read your text files. 

Args: 

doc_txt (str): Document as string. 

expression (str): Regular expression to find tokens. 

lower (boolean): If True, lowers all words. Defaults to True. 

simple (boolean): Uses simple regular expression (r'\w+'). Defaults to False. 

If set to True, argument `expression` will be ignored. 

Yields: 

Tokens 

Example: 

>>> list(tokenize("This is one example text.")) 

['this', 'is', 'one', 'example', 'text'] 

""" 

if lower: 

doc_txt = doc_txt.lower() 

if simple: 

pattern = regex.compile(r'\w+') 

else: 

pattern = regex.compile(expression) 

doc_txt = regex.sub("\.", "", doc_txt) 

doc_txt = regex.sub("‒", " ", doc_txt) 

doc_txt = regex.sub("–", " ", doc_txt) 

doc_txt = regex.sub("—", " ", doc_txt) 

doc_txt = regex.sub("―", " ", doc_txt) 

tokens = pattern.finditer(doc_txt) 

for match in tokens: 

yield match.group() 

def filter_pos_tags(doc_csv, pos_tags=['ADJ', 'V', 'NN']): 

"""Gets lemmas by selected POS-tags from DARIAH-DKPro-Wrapper output. 

Description: 

With this function you can select certain columns of a CSV file 

generated by `DARIAH-DKPro-Wrapper`_, a tool for natural language processing. 

Use the function `read_from_csv()` to read CSV files. 

.. _DARIAH-DKPro-Wrapper: 

https://github.com/DARIAH-DE/DARIAH-DKPro-Wrapper 

Args: 

doc_csv (DataFrame): DataFrame containing DARIAH-DKPro-Wrapper output. 

pos_tags (list[str]): List of DKPro POS-tags that should be selected. 

Defaults to '['ADJ', 'V', 'NN']'. 

Yields: 

Lemma. 

Example: 

>>> df = pd.DataFrame({'CPOS': ['CARD', 'ADJ', 'NN', 'NN'], 

... 'Lemma': ['one', 'more', 'example', 'text']}) 

>>> list(filter_pos_tags(df))[0] # doctest: +NORMALIZE_WHITESPACE 

1 more 

2 example 

3 text 

Name: Lemma, dtype: object 

""" 

log.info("Accessing %s ...", pos_tags) 

doc_csv = doc_csv[doc_csv['CPOS'].isin(pos_tags)] 

yield doc_csv['Lemma'] 

def split_paragraphs(doc_txt, sep=regex.compile('\n')): 

"""Splits the given document by paragraphs. 

Description: 

With this function you can split a document by paragraphs. You also have 

the ability to select a certain regular expression to split the document. 

Use the functions `read_from_txt()`, `read_from_tei()` or `read_from_csv()` 

to read your text files. 

Args: 

doc_txt (str): Document text. 

sep (regex.Regex): Separator indicating a paragraph. 

Returns: 

List of paragraphs. 

Example: 

>>> split_paragraphs("This test contains \\n paragraphs.") 

['This test contains ', ' paragraphs.'] 

""" 

if not hasattr(sep, 'match'): 

sep = regex.compile(sep) 

return sep.split(doc_txt) 

def segment_fuzzy(document, segment_size=5000, tolerance=0.05): 

"""Segments a document, tolerating existing chunks (like paragraphs). 

Description: 

Consider you have a document. You wish to split the document into 

segments of about 1000 tokens, but you prefer to keep paragraphs together 

if this does not increase or decrease the token size by more than 5%. 

Args: 

document: The document to process. This is an Iterable of chunks, each 

of which is an iterable of tokens. 

segment_size (int): The target length of each segment in tokens. 

tolerance (Number): How much may the actual segment size differ from 

the segment_size? If 0 < tolerance < 1, this is interpreted as a 

fraction of the segment_size, otherwise it is interpreted as an 

absolute number. If tolerance < 0, chunks are never split apart. 

Yields: 

Segments. Each segment is a list of chunks, each chunk is a list of 

tokens. 

Example: 

>>> list(segment_fuzzy([['This', 'test', 'is', 'very', 'clear'], 

... ['and', 'contains', 'chunks']], 2)) # doctest: +NORMALIZE_WHITESPACE 

[[['This', 'test']], 

[['is', 'very']], 

[['clear'], ['and']], 

[['contains', 'chunks']]] 

""" 

if tolerance > 0 and tolerance < 1: 

tolerance = round(segment_size * tolerance) 

current_segment = [] 

current_size = 0 

carry = None 

doc_iter = iter(document) 

try: 

while True: 

chunk = list(carry if carry else next(doc_iter)) 

carry = None 

current_segment.append(chunk) 

current_size += len(chunk) 

if current_size >= segment_size: 

too_long = current_size - segment_size 

too_short = segment_size - (current_size - len(chunk)) 

if tolerance >= 0 and min(too_long, too_short) > tolerance: 

chunk_part0 = chunk[:-too_long] 

carry = chunk[-too_long:] 

current_segment[-1] = chunk_part0 

elif too_long >= too_short: 

carry = current_segment.pop() 

yield current_segment 

current_segment = [] 

current_size = 0 

except StopIteration: 

pass 

# handle leftovers 

if current_segment: 

yield current_segment 

def segment(document, segment_size=1000, tolerance=0, chunker=None, 

tokenizer=None, flatten_chunks=False, materialize=False): 

"""Segments a document into segments of about `segment_size` tokens, respecting existing chunks. 

Description: 

Consider you have a document. You wish to split the document into 

segments of about 1000 tokens, but you prefer to keep paragraphs together 

if this does not increase or decrease the token size by more than 5%. 

This is a convenience wrapper around `segment_fuzzy()`. 

Args: 

segment_size (int): The target size of each segment, in tokens. 

tolerance (Number): see `segment_fuzzy` 

chunker (callable): a one-argument function that cuts the document into 

chunks. If this is present, it is called on the given document. 

tokenizer (callable): a one-argument function that tokenizes each chunk. 

flatten_chunks (bool): if True, undo the effect of the chunker by 

chaining the chunks in each segment, thus each segment consists of 

tokens. This can also be a one-argument function in order to 

customize the un-chunking. 

Example: 

>>> list(segment([['This', 'test', 'is', 'very', 'clear'], 

... ['and', 'contains', 'chunks']], 2)) # doctest: +NORMALIZE_WHITESPACE 

[[['This', 'test']], 

[['is', 'very']], 

[['clear'], ['and']], 

[['contains', 'chunks']]] 

""" 

if chunker is not None: 

document = chunker(document) 

if tokenizer is not None: 

document = map(tokenizer, document) 

segments = segment_fuzzy(document, segment_size, tolerance) 

if flatten_chunks: 

if not callable(flatten_chunks): 

def flatten_chunks(segment): 

return list(chain.from_iterable(segment)) 

segments = map(flatten_chunks, segments) 

if materialize: 

segments = list(segments) 

return segments 

def remove_features_from_file(doc_token_list, features_to_be_removed): 

"""Removes features using feature list. 

Description: 

With this function you can remove features from ppreprocessed files. 

Commit a list of features. 

Use the function `tokenize()` to access your files. 

Args: 

doc_token_list Union(list[str], str): List of all documents in the corpus 

and their tokens. 

features_to_be_removed list[str]: List of features that should be 

removed 

Yields: 

cleaned token array 

Todo: 

Example: 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> features_to_be_removed = ['example'] 

>>> test = remove_features_from_file(doc_tokens, features_to_be_removed) 

>>> list(test) 

[['short', 'text', 'text']] 

""" 

#log.info("Removing features ...") 

doc_token_array = np.array(doc_token_list) 

feature_array = np.array(features_to_be_removed) 

#get indices of features that should be deleted 

indices = np.where(np.in1d(doc_token_array, feature_array,)) 

doc_token_array = np.delete(doc_token_array, indices) 

yield doc_token_array.tolist() 

def create_mallet_import(doc_tokens_cleaned, doc_labels, outpath = os.path.join('tutorial_supplementals', 'mallet_input')): 

"""Creates files for mallet import. 

Description: 

With this function you can create preprocessed plain text files. 

Commit a list of full paths or one single path as argument. 

Use the function `remove_features_from_file()` to create a list of tokens 

per document. 

Args: 

doc_tokens_cleaned Union(list[str], str): List of tokens per document 

doc_labels list[str]: List of documents labels. 

Todo: 

Example: 

>>> doc_labels = ['examplefile'] 

>>> doc_tokens_cleaned = [['short', 'example', 'text']] 

>>> create_mallet_import(doc_tokens_cleaned, doc_labels) 

>>> outpath = os.path.join('tutorial_supplementals', 'mallet_input') 

>>> os.path.isfile(os.path.join(outpath, 'examplefile.txt')) 

True 

""" 

#log.info("Generating mallet input files ...") 

if not os.path.exists(outpath): 

os.makedirs(outpath) 

for tokens, label in zip(doc_tokens_cleaned, doc_labels): 

with open(os.path.join(outpath,label+'.txt'), 'w', encoding="utf-8") as f: 

for token in tokens: 

f.write(' '.join(token)) 

def create_doc_term_matrix(tokens, doc_labels): 

"""Creates a document-term matrix 

Description: 

With this function you can create a document-term matrix 

where rows correspond to documents in the collection and columns  

correspond to terms. 

Use the function `tokenize()` to tokenize your text files and 

Use the function `_wordcounts()` to generate the wordcounts 

Args: 

doc_labels (list[str]): List of doc labels as string 

tokens (list): List of tokens. 

Returns: 

DataFrame. 

Example: 

>>> example = create_doc_term_matrix('example', 'label') 

>>> print(isinstance(example, pd.DataFrame)) 

>>> True 

""" 

df = pd.DataFrame([_wordcounts(doc, label) for doc, label in zip(tokens, doc_labels)]) 

df = df.fillna(0) 

return df.loc[:, df.sum().sort_values(ascending=False).index] 

def _wordcounts(doc, label): 

"""Creates a Series with wordcounts 

Description: 

Only the function 'create_doc_term_matrix() uses this private  

function.  

Args: 

doc (list[tokens]): List of tokens 

label (String): String with document_label. 

Returns: 

Pandas Series. 

ToDo: 

Complete documetation 

Example: 

""" 

s = pd.Series(Counter(doc)) 

s.name = label 

return s 

def create_dictionary(tokens): 

"""Creates a dictionary of unique tokens with identifier. 

Description: 

With this function you can create a dictionary of unique tokens as key 

and an identifier as value. 

Use the function `tokenize()` to tokenize your text files. 

Args: 

tokens (list): List of tokens. 

Returns: 

Dictionary. 

Example: 

>>> create_dictionary(['example']) 

{'example': 1} 

""" 

if all(isinstance(element, list) for element in tokens): 

tokens = {token for element in tokens for token in element} 

return {token: id_ for id_, token in enumerate(set(tokens), 1)} 

def _create_large_counter(doc_labels, doc_tokens, type_dictionary): 

"""Creates a dictionary of dictionaries. 

Description: 

Only the function `create_sparse_bow()` uses this private function to 

create a dictionary of dictionaries. 

The first level consists of the document label as key, and the dictionary 

of counts as value. The second level consists of token ID as key, and the 

count of tokens in document pairs as value. 

Args: 

doc_labels (list): List of doc labels. 

doc_tokens (list): List of tokens. 

type_dictionary (dict): Dictionary of {token: id}. 

Returns: 

Dictionary of dictionaries. 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> type_dictionary = {'short': 1, 'example': 2, 'text': 3} 

>>> isinstance(_create_large_counter(doc_labels, doc_tokens, type_dictionary), defaultdict) 

True 

""" 

largecounter = defaultdict(dict) 

for doc, tokens in zip(doc_labels, doc_tokens): 

largecounter[doc] = Counter( 

[type_dictionary[token] for token in tokens]) 

return largecounter 

def _create_sparse_index(largecounter): 

"""Creates a sparse index for pandas DataFrame. 

Description: 

Only the function `create_sparse_bow()` uses this private function to 

create a pandas multiindex out of tuples. 

The multiindex represents document ID to token IDs relations. 

Args: 

largecounter (dict): Dictionary of {document: {token: frequency}}. 

Returns: 

Pandas MultiIndex. 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> type_dictionary = {'short': 1, 'example': 2, 'text': 3} 

>>> largecounter = _create_large_counter(doc_labels, doc_tokens, type_dictionary) 

>>> isinstance(_create_sparse_index(largecounter), pd.MultiIndex) 

True 

""" 

tuples = [] 

for key in range(1, len(largecounter) + 1): 

if len(largecounter[key]) == 0: 

tuples.append((key, 0)) 

for value in largecounter[key]: 

tuples.append((key, value)) 

sparse_index = pd.MultiIndex.from_tuples( 

tuples, names=['doc_id', 'token_id']) 

return sparse_index 

def create_sparse_bow(doc_labels, doc_tokens, type_dictionary, doc_dictionary): 

"""Creates sparse matrix for bag-of-words model. 

Description: 

This function creates a sparse DataFrame ('bow' means `bag-of-words`_) 

containing document and type identifier as multiindex and type 

frequencies as values representing the counts of tokens for each token 

in each document. 

It is also the main function that incorporates the private functions 

`_create_large_counter()` and `_create_sparse_index()``. 

Use the function `get_labels()` for `doc_labels`, `tokenize()` for 

`doc_tokens`, and `create_dictionary()` for `type_dictionary` as well 

as for `doc_ids`. 

Use the function `create_dictionary()` to generate the dictionaries 

`type_dictionary` and `doc_dictionary`. 

.. _bag-of-words: 

https://en.wikipedia.org/wiki/Bag-of-words_model 

Args: 

doc_labels (list[str]): List of doc labels as string. 

doc_tokens (list[str]): List of tokens as string. 

type_dictionary (dict[str]): Dictionary with {token: id}. 

doc_ids (dict[str]): Dictionary with {document label: id}. 

Returns: 

Multiindexed Pandas DataFrame. 

ToDo: 

* Test if it's necessary to build sparse_df_filled with int8 zeroes instead of int64. 

* Avoid saving sparse bow as .mm file to ingest into gensim. 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'text']] 

>>> type_dictionary = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> len(create_sparse_bow(doc_labels, doc_tokens, type_dictionary, doc_ids)) 

3 

""" 

temp_counter = _create_large_counter( 

doc_labels, doc_tokens, type_dictionary) 

largecounter = {doc_dictionary[key]: value for key, value in temp_counter.items()} 

sparse_index = _create_sparse_index(largecounter) 

sparse_bow_filled = pd.DataFrame( 

np.zeros((len(sparse_index), 1), dtype=int), index=sparse_index) 

index_iterator = sparse_index.groupby( 

sparse_index.get_level_values('doc_id')) 

for doc_id in range(1, len(sparse_index.levels[0]) + 1): 

for token_id in [val[1] for val in index_iterator[doc_id]]: 

sparse_bow_filled.set_value( 

(doc_id, token_id), 0, int(largecounter[doc_id][token_id])) 

return sparse_bow_filled 

def save_sparse_bow(sparse_bow, output): 

"""Saves sparse matrix for bag-of-words model. 

Description: 

With this function you can save the sparse matrix as `.mm file`_. 

.. _.mm file: http://math.nist.gov/MatrixMarket/formats.html#MMformat 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

output (str): Path to output file without extension, e.g. /tmp/sparsebow. 

Returns: 

None. 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'text']] 

>>> type_dictionary = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, type_dictionary, doc_ids) 

>>> save_sparse_bow(sparse_bow, 'sparsebow') 

>>> import os.path 

>>> os.path.isfile('sparsebow.mm') 

True 

""" 

num_docs = sparse_bow.index.get_level_values("doc_id").max() 

num_types = sparse_bow.index.get_level_values("token_id").max() 

sum_counts = sparse_bow[0].sum() 

header_string = str(num_docs) + " " + str(num_types) + \ 

" " + str(sum_counts) + "\n" 

with open('.'.join([output, 'mm']), 'w', encoding="utf-8") as f: 

f.write("%%MatrixMarket matrix coordinate real general\n") 

f.write(header_string) 

sparse_bow.to_csv(f, sep=' ', header=None) 

def find_stopwords(df, mfw=100, id_types=None): 

"""Creates a stopword list. 

Description: 

With this function you can determine most frequent words, also known as 

stopwords. First, you have to translate your corpus into the bag-of-words 

model using the function `create_sparse_matrix()` and create an dictionary 

containing types and identifier using `create_dictionary()`. 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

id_types (dict[str]): Dictionary with {token: id}. 

mfw (int): Target size of most frequent words to be considered. 

Returns: 

Most frequent words in a list. 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'short', 'example', 'text']] 

>>> id_types = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, id_types, doc_ids) 

>>> find_stopwords(sparse_bow, 1, id_types) 

['short'] 

""" 

log.info("Finding stopwords ...") 

if isinstance(df.index, pd.MultiIndex): 

type2id = {value: key for key, value in id_types.items()} 

sparse_bow_collapsed = df.groupby( 

df.index.get_level_values('token_id')).sum() 

sparse_bow_stopwords = sparse_bow_collapsed[0].nlargest(mfw) 

stopwords = [type2id[key] 

for key in sparse_bow_stopwords.index.get_level_values('token_id')] 

return stopwords 

else: 

return df.iloc[:,:mfw].columns.tolist() 

def find_hapax(df, id_types=None): 

"""Creates a list with hapax legommena. 

Description: 

With this function you can determine hapax legomena for each document. 

First, you have to translate your corpus into the bag-of-words 

model using the function `create_sparse_matrix()` and create an dictionary 

containing types and identifier using `create_dictionary()`. 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

id_types (dict[str]): Dictionary with {token: id}. 

Returns: 

Hapax legomena in a list. 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> id_types = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, id_types, doc_ids) 

>>> find_hapax(sparse_bow, id_types) 

['short'] 

""" 

log.info("Finding hapax legomena ...") 

if isinstance(df.index, pd.MultiIndex): 

type2id = {value: key for key, value in id_types.items()} 

sparse_bow_collapsed = df.groupby( 

df.index.get_level_values('token_id')).sum() 

sparse_bow_hapax = sparse_bow_collapsed.loc[sparse_bow_collapsed[0] == 1] 

hapax = [type2id[key] 

for key in sparse_bow_hapax.index.get_level_values('token_id')] 

return hapax 

else: 

#return df.loc[:,(df.isin([1])).any()].columns.tolist() 

return df.loc[:, df.max() == 1].columns.tolist() 

def remove_features_from_df(df, features, id_types=None): 

"""Removes features based on a list of words (types). 

Description: 

With this function you can clean your corpus from stopwords and hapax 

legomena. 

First, you have to translate your corpus into the bag-of-words 

model using the function `create_sparse_bow()` and create a dictionary 

containing types and identifier using `create_dictionary()`. 

Use the functions `find_stopwords()` and `find_hapax()` to generate a 

feature list. 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

features Union(set, list): Set or list containing features to remove. 

(not included) features (str): Text as iterable. 

Returns: 

Clean corpus. 

ToDo: 

* Adapt function to work with mm-corpus format. 

Example: 

>>> doc_labels = ['exampletext'] 

>>> doc_tokens = [['short', 'example', 'example', 'text', 'text']] 

>>> id_types = {'short': 1, 'example': 2, 'text': 3} 

>>> doc_ids = {'exampletext': 1} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, id_types, doc_ids) 

>>> features = ['short'] 

>>> len(remove_features(sparse_bow, features, id_types)) 

2 

""" 

log.info("Removing features ...") 

if isinstance(df.index, pd.MultiIndex): 

if isinstance(features, list): 

features = set(features) 

stoplist_applied = [word for word in set(id_types.keys()) if word in features] 

clean_df = df.drop([id_types[word] for word in stoplist_applied], level='token_id') 

return clean_df 

else: 

features = [token for token in features if token in df.columns] 

df.drop(features, inplace=True, axis=1) 

return df 

def make_doc2bow_list(sparse_bow): 

"""Creates doc2bow_list for gensim. 

Description: 

With this function you can create a doc2bow_list as input for the gensim 

function `get_document_topics()` to show topics for each document. 

Args: 

sparse_bow (DataFrame): DataFrame with term and term frequency by document. 

Returns: 

List of lists containing tuples. 

Example: 

>>> doc_labels = ['exampletext1', 'exampletext2'] 

>>> doc_tokens = [['test', 'corpus'], ['for', 'testing']] 

>>> type_dictionary = {'test': 1, 'corpus': 2, 'for': 3, 'testing': 4} 

>>> doc_dictionary = {'exampletext1': 1, 'exampletext2': 2} 

>>> sparse_bow = create_sparse_bow(doc_labels, doc_tokens, type_dictionary, doc_dictionary) 

>>> from gensim.models import LdaModel 

>>> from gensim.corpora import Dictionary 

>>> corpus = [['test', 'corpus'], ['for', 'testing']] 

>>> dictionary = Dictionary(corpus) 

>>> documents = [dictionary.doc2bow(document) for document in corpus] 

>>> model = LdaModel(corpus=documents, id2word=dictionary, iterations=1, passes=1, num_topics=1) 

>>> make_doc2bow_list(sparse_bow) 

[[(1, 1), (2, 1)], [(3, 1), (4, 1)]] 

""" 

doc2bow_list = [] 

for doc in sparse_bow.index.groupby(sparse_bow.index.get_level_values('doc_id')): 

temp = [(token, count) for token, count in zip( 

sparse_bow.loc[doc].index, sparse_bow.loc[doc][0])] 

doc2bow_list.append(temp) 

return doc2bow_list 

def lda2dataframe(model, vocab, num_keys=10): 

"""Converts lda output to a DataFrame 

Description: 

With this function you can convert lda output to a DataFrame,  

a more convenient datastructure. 

Note: 

Args: 

model: LDA model. 

vocab (list[str]): List of strings containing corpus vocabulary.  

num_keys (int): Number of top keywords for topic 

Returns: 

DataFrame 

Example: 

>>> import lda 

>>> corpus = [['test', 'corpus'], ['for', 'testing']] 

>>> doc_term_matrix = create_doc_term_matrix(corpus, ['doc1', 'doc2']) 

>>> vocab = doc_term_matrix.columns 

>>> model = lda.LDA(n_topics=1, n_iter=1) 

>>> model.fit(doc_term_matrix.as_matrix().astype(int)) 

>>> df = lda2dataframe(model, vocab, num_keys=1) 

>>> len(df) == 1 

True 

""" 

topics = [] 

topic_word = model.topic_word_ 

for i, topic_dist in enumerate(topic_word): 

topics.append(np.array(vocab)[np.argsort(topic_dist)][:-num_keys-1:-1]) 

return pd.DataFrame(topics, index=['Topic ' + str(n+1) for n in range(len(topics))], columns=['Key ' + str(n+1) for n in range(num_keys)]) 

def gensim2dataframe(model, num_keys=10): 

"""Converts gensim output to DataFrame. 

Description: 

With this function you can convert gensim output (usually a list of 

tuples) to a DataFrame, a more convenient datastructure. 

Args: 

model: Gensim LDA model. 

num_keys (int): Number of top keywords for topic. 

Returns: 

DataFrame. 

ToDo: 

Example: 

>>> from gensim.models import LdaModel 

>>> from gensim.corpora import Dictionary 

>>> corpus = [['test', 'corpus'], ['for', 'testing']] 

>>> dictionary = Dictionary(corpus) 

>>> documents = [dictionary.doc2bow(document) for document in corpus] 

>>> model = LdaModel(corpus=documents, id2word=dictionary, iterations=1, passes=1, num_topics=1) 

>>> isinstance(gensim2dataframe(model, 4), pd.DataFrame) 

True 

""" 

num_topics = model.num_topics 

topics_df = pd.DataFrame(index = range(num_topics), 

columns= range(num_keys)) 

topics = model.show_topics(num_topics = model.num_topics, formatted=False) 

for topic, values in topics: 

keyword = [value[0] for value in values] 

topics_df.loc[topic] = keyword 

return topics_df 

def lda_doc_topic(model, topics, doc_labels): 

"""Creates a doc_topic_matrix for lda output. 

Description: 

With this function you can convert lda output to a DataFrame,  

a more convenient datastructure. 

Use 'lda2DataFrame()' to get topics. 

Note: 

Args: 

model: Gensim LDA model. 

topics: DataFrame. 

doc_labels (list[str]): List of doc labels as string. 

Returns: 

DataFrame 

Example: 

>>> import lda 

>>> corpus = [['test', 'corpus'], ['for', 'testing']] 

>>> doc_term_matrix = create_doc_term_matrix(corpus, ['doc1', 'doc2']) 

>>> vocab = doc_term_matrix.columns 

>>> model = lda.LDA(n_topics=1, n_iter=1) 

>>> model.fit(doc_term_matrix.as_matrix().astype(int)) 

>>> topics = lda2dataframe(model, vocab) 

>>> doc_topic = lda_doc_topic(model, vocab, ['doc1', 'doc2']) 

>>> len(doc_topic.T) == 2 

True 

""" 

topic_labels = [] 

topic_terms = [x[:3] for x in topics.values.tolist()] 

for topic in topic_terms: 

topic_labels.append(" ".join(topic)) 

df = pd.DataFrame(model.doc_topic_).T 

df.columns = doc_labels 

df.index = topic_labels 

return df.reindex_axis(sorted(df.columns), axis=1)