-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathcompute_fever_wordstats.py
212 lines (165 loc) · 8.45 KB
/
compute_fever_wordstats.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
from os import listdir
import math
import os
from os.path import isfile, join
import jsonlines
from tqdm import tqdm
import nltk
from nltk.tokenize import word_tokenize
from nltk.util import ngrams
import unicodedata
import numpy as np
import random
import csv
import pickle
import pandas as pd
from collections import defaultdict, Counter
from scipy.stats import pearsonr, spearmanr
from nltk.tokenize import RegexpTokenizer
valid_file = 'fever.dev.jsonl'
train_file = 'fever.train.jsonl'
TOP_N = 20
MIN_FREQ = 5
NGRAM = 2
def get_single_stopwords(dataset, ngram=1):
tokenizer = RegexpTokenizer(r'\w+')
fp = open(dataset, "r", encoding='utf-8')
reader = jsonlines.Reader(fp)
global_word_counter = defaultdict(int)
phrases = 0
for dictionary in tqdm(reader.iter()):
claim = dictionary['claim']
words = tokenizer.tokenize(claim.lower())
for i in range(len(words) + 1):
if i > ngram - 1:
phrase = ' '.join(words[i - ngram:i])
global_word_counter[phrase] += 1
counter = Counter(global_word_counter)
stop_words = counter.most_common(10)
print (sum(counter.values()))
stop_words = [word[0] for word in stop_words]
print (stop_words)
return stop_words
def get_counters(dataset, ngram=NGRAM):
stop_words = get_single_stopwords(train_file, ngram=1)
tokenizer = RegexpTokenizer(r'\w+')
fp = open(dataset, "r", encoding='utf-8')
reader = jsonlines.Reader(fp)
global_word_counter = defaultdict(int)
global_label_counter = defaultdict(int)
phrases = 0
label_word_counter = defaultdict(lambda: defaultdict(int))
for dictionary in tqdm(reader.iter()):
label = dictionary['label']
claim = dictionary['claim']
#words = word_tokenize(claim.lower())
words = tokenizer.tokenize(claim.lower())
words = [words[i] for i in range(len(words)) if words[i] not in stop_words]
bigrams = ngrams(words, NGRAM)
"""
for word in words:
global_word_counter[word] += 1
global_label_counter[label] += 1
label_word_counter[label][word] += 1
phrases += 1
"""
for bigram in bigrams:
bigram = ' '.join(bigram)
global_word_counter[bigram] += 1
global_label_counter[label] += 1
label_word_counter[label][bigram] += 1
phrases += 1
"""
for i in range(len(words) + 1):
if i > ngram - 1:
phrase = ' '.join(words[i - ngram:i])
global_word_counter[phrase] += 1
global_label_counter[label] += 1
label_word_counter[label][phrase] += 1
phrases += 1
"""
print ('Total count: ' + str(phrases))
return global_word_counter, label_word_counter, global_label_counter, phrases
valid_global_word_counter, valid_label_word_counter, valid_global_label_counter, valid_words = get_counters(valid_file)
train_global_word_counter, train_label_word_counter, train_global_label_counter, train_words = get_counters(train_file)
corr = {'SUPPORTS': [], 'REFUTES': [], 'NOT ENOUGH INFO': []}
for label in train_label_word_counter.keys():
words = []
scores = []
pmis = []
valid_pmis = []
valid_scores = []
freqs = []
valid_freqs = []
p_l_train = train_global_label_counter[label] / train_words
p_l_valid = valid_global_label_counter[label] / valid_words
print (train_words)
word_counter = train_label_word_counter[label]
for w in word_counter:
if train_global_word_counter[w] < MIN_FREQ:
continue
# p(label | word)
score = word_counter[w] / train_global_word_counter[w]
pmi = math.log(score / p_l_train)
#pmi = max(0, pmi)
if w in valid_global_word_counter:
valid_score = valid_label_word_counter[label][w] / valid_global_word_counter[w]
if valid_score == 0:
valid_pmi = float('inf')
else:
valid_pmi = math.log(valid_score / p_l_valid)
#valid_pmi = max(0, math.log(valid_score / p_l_valid))
else:
valid_score = 0
valid_pmi = float('inf')
words.append(w)
scores.append(score)
pmis.append(pmi)
freqs.append(word_counter[w])
valid_freqs.append(valid_label_word_counter[label][w])
valid_scores.append(valid_score)
valid_pmis.append(valid_pmi)
assert(len(words) == len(scores) == len(freqs) == len(valid_freqs) == len(valid_scores) == len(pmis))
pmis_x_freq = list(np.array(pmis)*freqs/train_words)
valid_pmis_x_freq = list(np.array(valid_pmis)*valid_freqs/valid_words)
pmis_x_freq, pmis, scores, freqs, words, valid_scores, valid_pmis, valid_pmis_x_freq, valid_freqs = (list(t) for t in zip(*sorted(zip(pmis_x_freq, pmis, scores, freqs, words, valid_scores, valid_pmis, valid_pmis_x_freq, valid_freqs), reverse=True)))
print("")
print("---- {}".format(label))
print("{:20} | {:7} | {:5} | {:7} | {:5}".format('word', 'lmi', 'p(l|w)', 'valid_lmi', 'valid_p(l|w)'))
#print("{:20} | {:6} | {:7} | {:7} | {:4} | {:11} | {:10} | {:10} | {:10}".format('word', 'score', 'pmi', 'lmi', 'freq', 'valid score', 'valid_pmi', 'valid_lmi', 'valid freq'))
print ("-"*80)
#filepath = 'top_20_lmi_p_2_' + label + '.csv'
filepath = 'top_1000_unigram_' + label + '.csv'
with open(filepath, 'w') as f:
csv_writer = csv.writer(f, delimiter=',', quotechar='"', quoting=csv.QUOTE_MINIMAL)
for i in range(min(TOP_N, len(words))):
#print("{:20} | {:6} | {:7} | {:7} | {:4} | {:11} | {:10} | {:10} | {:10}".format(words[i], round(scores[i], 3), round(pmis[i],3), round(pmis_x_freq[i],3), freqs[i], round(valid_scores[i],3), round(valid_pmis[i],3), round(valid_pmis_x_freq[i],3), valid_freqs[i]))
if not math.isnan(valid_pmis_x_freq[i]):
print("{:20} | {:7} | {:5} | {:7} | {:5}".format(words[i], int(round(pmis_x_freq[i]*10**6)), round(scores[i],2), int(round(valid_pmis_x_freq[i]*10**6)), round(valid_scores[i],2)))
csv_writer.writerow([words[i], int(round(pmis_x_freq[i]*10**6)), round(scores[i],2), freqs[i], int(round(valid_pmis_x_freq[i]*10**6)), round(valid_scores[i],2), valid_freqs[i]])
else:
print("{:20} | {:7} | {:5} | {:7} | {:5}".format(words[i], int(round(pmis_x_freq[i]*10**6)), round(scores[i],2), valid_pmis_x_freq[i], round(valid_scores[i],2)))
csv_writer.writerow([words[i], int(round(pmis_x_freq[i]*10**6)), round(scores[i],2), freqs[i], valid_pmis_x_freq[i], round(valid_scores[i],2), valid_freqs[i]])
'''
extra_words = ['did not', 'yet to', 'does not', 'refused to', 'failed to', 'unable to', 'incapable being', 'united states', 'least one', 'at least', 'person who', 'stars actor', 'least one', 'won award', 'played for']
#extra_words = ['at least one']
for w in extra_words:
i = words.index(w)
if not math.isnan(valid_pmis_x_freq[i]):
print("{:20} | {:7} | {:5} | {:7} | {:5}".format(words[i], int(round(pmis_x_freq[i]*10**6)), round(scores[i],2), int(round(valid_pmis_x_freq[i]*10**6)), round(valid_scores[i],2)))
csv_writer.writerow([words[i], int(round(pmis_x_freq[i]*10**6)), round(scores[i],2), freqs[i], int(round(valid_pmis_x_freq[i]*10**6)), round(valid_scores[i],2), valid_freqs[i]])
else:
print("{:20} | {:7} | {:5} | {:7} | {:5}".format(words[i], int(round(pmis_x_freq[i]*10**6)), round(scores[i],2), valid_pmis_x_freq[i], round(valid_scores[i],2)))
csv_writer.writerow([words[i], int(round(pmis_x_freq[i]*10**6)), round(scores[i],2), freqs[i], valid_pmis_x_freq[i], round(valid_scores[i],2), valid_freqs[i]])
'''
limits = [10, 20, 50, 100, 200, 500, 1000]
#corr_filepath = 'corr_unigram_1000.pkl'
corr_ind = []
for limit in limits:
pears = pearsonr(scores[0:limit], valid_scores[0:limit])
print ("pearson correlation for top {}: {} (p-value: {})".format(limit, round(pears[0],3), round(pears[1],3)))
corr_ind.append(round(pears[0],3))
spear = spearmanr(scores[0:limit], valid_scores[0:limit])
print ("spearman correlation for top {}: {} (p-value: {})".format(limit, round(spear[0],3), round(spear[1],3)))
corr[label] = [limits, corr_ind]
#pickle.dump(corr, open(corr_filepath, 'wb'))