Pinpoint-Web / Pinpoint_Internal /Aggregator_TfIdf.py
James Stevenson
added lib
246df79
raw
history blame
1.49 kB
from sklearn.feature_extraction.text import TfidfVectorizer
from Pinpoint_Internal.Logger import *
class tf_idf_aggregator():
"""
A wrapper class around SKlearn for retrieving TF-IDF scores.
"""
def get_tf_idf_scores(self, ngrams_vocabulary, corpus_data=None, file_name_to_read=None):
"""
Used to generate a TF IDF score based of a vocabulary of Ngrams and a data corpus.
:param ngrams_vocabulary:
:param corpus_data:
:param file_name_to_read:
:return: a dictionary of the pairing name and their score
"""
logger.print_message("Getting TF IDF scores")
if corpus_data is None and file_name_to_read is None:
raise Exception("No data supplied to retrieve n_grams")
if corpus_data is None and file_name_to_read is not None:
with open(file_name_to_read, 'r') as file_to_read:
corpus_data = file_to_read.read()
tfidf = TfidfVectorizer(vocabulary=ngrams_vocabulary, stop_words='english', ngram_range=(1, 2))
tfs = tfidf.fit_transform([corpus_data])
feature_names = tfidf.get_feature_names()
corpus_index = [n for n in corpus_data]
rows, cols = tfs.nonzero()
dict_of_scores = {}
for row, col in zip(rows, cols):
dict_of_scores[feature_names[col]] = tfs[row, col]
logger.print_message((feature_names[col], corpus_index[row]), tfs[row, col])
return dict_of_scores