-
Notifications
You must be signed in to change notification settings - Fork 3
/
log_odds_ratio.py
217 lines (189 loc) · 8.76 KB
/
log_odds_ratio.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
#####################################################################
# LogOddsRatio Class
#
# A class for computing Log-odds-ratio with informative Dirichlet priors
#
# See http://languagelog.ldc.upenn.edu/myl/Monroe.pdf for more detail
#
#####################################################################
__author__ = "Kornraphop Kawintiranon"
__email__ = "kornraphop.k@gmail.com"
import math
from loguru import logger
import tqdm
import numpy as np
import pandas as pd
import argparse
from nltk.corpus import stopwords
from nltk.tokenize import TweetTokenizer
from text_helper import *
class LogOddsRatio:
"""
Log-odds-ratio with informative Dirichlet priors
"""
def __init__(self, corpus_i, corpus_j, background_corpus=None, lower_case=True, rm_stopwords=True, rm_punctuations=True, tokenizer=None):
"""
Create a class object and prepare word counts for log-odds-ratio computation
Args:
corpus_i:
A list of documents, each contains a string
corpus_j:
A list of documents, each contains a string
background_corpus (default = None):
If None, it will be assigned to a concatenation of `corpus_i` and `corpus_j`
lower_case:
Whether lower case all words
rm_stopwords:
Whether remove stopwords in preprocessing step
rm_punctuations:
Whether remove punctuations in preprocessing step
tokenizer:
To specify a specific tokenizer for tokenization step
"""
def preprocessing(corpus):
if lower_case:
corpus = [text.lower() for text in corpus]
corpus = decontract(corpus)
tokenized_corpus = parallel_tokenize(corpus, tokenizer)
if rm_stopwords:
tokenized_corpus = remove_stopwords(tokenized_corpus)
if rm_punctuations:
tokenized_corpus = remove_punctuations(tokenized_corpus)
return tokenized_corpus
# Convert a list of string into a list of lists of words
logger.info("Preprocessing corpus-i")
corpus_i = preprocessing(corpus_i)
logger.info("Preprocessing corpus-j")
corpus_j = preprocessing(corpus_j)
if background_corpus != None:
logger.info("Preprocessing corpus-background")
background_corpus = preprocessing(background_corpus)
# Compute word counts of every words on each corpus separately
logger.info("Getting word counts from corpus-i")
self.y_i = get_word_counts(corpus_i)
logger.info("Getting word counts from corpus-j")
self.y_j = get_word_counts(corpus_j)
logger.info("Getting word counts from corpus-background")
if background_corpus:
self.alpha = get_word_counts(background_corpus)
else:
# Combine words and sum their counts of corpus i and j in case no specified background corpus
self.alpha = {k: self.y_i.get(k, 0) + self.y_j.get(k, 0) for k in set(self.y_i) | set(self.y_j)}
# Sort dicts
logger.debug("Start sorting and backing up to files")
self.y_i = {k: v for k, v in sorted(self.y_i.items(), key=lambda item: item[1], reverse=True)}
self.y_j = {k: v for k, v in sorted(self.y_j.items(), key=lambda item: item[1], reverse=True)}
self.alpha = {k: v for k, v in sorted(self.alpha.items(), key=lambda item: item[1], reverse=True)}
# Write to files as backup
with open("vocabs_i.txt", "w") as f:
for k, v in self.y_i.items():
f.write(f"{k},{v}\n")
with open("vocabs_j.txt", "w") as f:
for k, v in self.y_j.items():
f.write(f"{k},{v}\n")
with open("vocabs_alpha.txt", "w") as f:
for k, v in self.alpha.items():
f.write(f"{k},{v}\n")
# Initialize necessary variables
self.delta = None
self.sigma_2 = None
self.z_scores = None
# Compute
logger.info("Start computing delta")
self._compute_delta()
logger.info("Start computing sigma^2")
self._compute_sigma_2()
logger.info("Start computing Z-score")
self._compute_z_scores()
# Sort dicts
logger.debug("Start sorting and backing up to files")
self.delta = {k: v for k, v in sorted(self.delta.items(), key=lambda item: item[1], reverse=True)}
self.sigma_2 = {k: v for k, v in sorted(self.sigma_2.items(), key=lambda item: item[1], reverse=True)}
self.z_scores = {k: v for k, v in sorted(self.z_scores.items(), key=lambda item: item[1], reverse=True)}
# Write to files as backup
with open("delta.txt", "w") as f:
for k, v in self.delta.items():
f.write(f"{k},{v}\n")
with open("sigma_2.txt", "w") as f:
for k, v in self.sigma_2.items():
f.write(f"{k},{v}\n")
with open("z_scores.txt", "w") as f:
for k, v in self.z_scores.items():
f.write(f"{k},{v}\n")
def _compute_delta(self):
""" The usage difference for word w among two corpora i and j
"""
self.delta = dict()
n_i = sum(self.y_i.values())
n_j = sum(self.y_j.values())
alpha_zero = sum(self.alpha.values())
logger.debug(f"Size of corpus-i: {n_i}")
logger.debug(f"Size of corpus-j: {n_j}")
logger.debug(f"Size of background corpus: {alpha_zero}")
try:
for w in set(self.y_i) | set(self.y_j): # iterate through all words among two corpora
first_log = math.log10((self.y_i.get(w, 0) + self.alpha.get(w, 0)) / (n_i + alpha_zero - self.y_i.get(w, 0) - self.alpha.get(w, 0)))
second_log = math.log10((self.y_j.get(w, 0) + self.alpha.get(w, 0)) / (n_j + alpha_zero - self.y_j.get(w, 0) - self.alpha.get(w, 0)))
self.delta[w] = first_log - second_log
except ValueError as e:
logger.debug(f"Y-i of the word {w}:", self.y_i.get(w, 0))
logger.debug(f"alpha of the word {w}:", self.alpha.get(w, 0))
logger.debug(f"value:", (self.y_i.get(w, 0) + self.alpha.get(w, 0)) /
(n_i + alpha_zero - self.y_i.get(w, 0) - self.alpha.get(w, 0)))
raise e
def _compute_sigma_2(self):
""" Compute estimated values of sigma squared
"""
self.sigma_2 = dict()
for w in self.delta:
self.sigma_2[w] = (1 / (self.y_i.get(w, 0) + self.alpha.get(w, 0))) + (1 / (self.y_j.get(w, 0) + self.alpha.get(w, 0)))
def _compute_z_scores(self):
self.z_scores = dict()
for w in self.delta:
self.z_scores[w] = self.delta.get(w, 0) / math.sqrt(self.sigma_2.get(w, 0))
def main():
# Argument setup
parser = argparse.ArgumentParser()
parser.add_argument("--filepath_corpus_i", type=str, required=True)
parser.add_argument("--filepath_corpus_j", type=str, required=True)
parser.add_argument("--filepath_background_corpus", default=None, type=str, required=False)
parser.add_argument("--save_top_words", default=None, type=int, required=False)
parser.add_argument("--log_level", default=None, type=str, required=False)
args = parser.parse_args()
# Set new log level, default is "DEBUG"
if args.log_level != None:
logger.remove()
logger.add(sys.stderr, level=args.log_level)
# Read file into list of texts
df_corpus_i = pd.read_csv(args.filepath_corpus_i)
corpus_i = [text.strip() for text in df_corpus_i["text"]]
del df_corpus_i
df_corpus_j = pd.read_csv(args.filepath_corpus_j)
corpus_j = [text.strip() for text in df_corpus_j["text"]]
del df_corpus_j
if args.filepath_background_corpus != None:
df_corpus_bg = pd.read_csv(args.filepath_background_corpus)
corpus_bg = [text.strip() for text in df_corpus_bg["text"]]
del df_corpus_bg
else:
corpus_bg = None
# Specify the tweet tokenizer
tweet_tokenizer = TweetTokenizer()
# Start log-odds-ratio preprocessing
log_odds_ratio = LogOddsRatio(
corpus_i, corpus_j, corpus_bg, tokenizer=tweet_tokenizer)
# Save top words into a file
if args.save_top_words != None and args.save_top_words > 0:
if args.save_top_words > len(log_odds_ratio.z_scores):
raise ValueError("--save_top_words must be less than or equal to vocab size")
logger.info(f"Saving top {args.save_top_words} words ranked by Z-score")
with open("top_words.txt", "w") as f:
i = 0
for k, v in log_odds_ratio.z_scores.items():
f.write(k+"\n")
i += 1
if i >= args.save_top_words:
break
if __name__ == "__main__":
main()
logger.success("Done!")