-
Notifications
You must be signed in to change notification settings - Fork 0
/
sstb_process_data.py
206 lines (179 loc) · 6.92 KB
/
sstb_process_data.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
import numpy as np
import cPickle
from collections import defaultdict
import re
import pandas as pd
from nltk.tag import StanfordPOSTagger
import csv
import sys
def get_split_num(split):
if split == 'train' or split == 'train_phrases':
return 0
elif split == 'test':
return 1
elif split == 'dev':
return 2
return -1
def sentiment_label_for_binary(sentiment):
if sentiment < 2:
return 0 # negative
elif sentiment > 2:
return 1 # positive
else:
print "invalid sentiment for binary case"
sys.exit()
def build_data_cv(data_file, all_phrases, binary, min_len=4):
revs = []
vocab = defaultdict(float)
pos_vocab = defaultdict(float)
pos_tagger = StanfordPOSTagger(
'pos-tag/english-left3words-distsim.tagger',
'pos-tag/stanford-postagger.jar',
'utf8', False, '-mx2000m')
splits = ['train', 'test', 'dev']
sentence_set = set()
for split in splits:
with open(data_file.format(split), "rb") as f:
reader = csv.reader(f)
revs_text = []
sents = []
for row in reader:
rev, sent = row[0], int(row[1])
if binary and sent == 2: # skip neutral if binary
continue
rev = clean_str_sst(rev)
if split == 'train':
sentence_set.add(rev)
rev_tokens = rev.split()
revs_text.append(rev_tokens)
sent = sentiment_label_for_binary(sent) if binary else sent # check for binary case
sents.append(sent)
revs_tagged = pos_tagger.tag_sents(revs_text)
for i in range(len(revs_tagged)):
rev_tagged = revs_tagged[i]
text = list(zip(*rev_tagged)[0])
tag = list(zip(*rev_tagged)[1])
for word in set(text):
vocab[word] += 1
for postag in set(tag):
pos_vocab[postag] += 1
rev_datum = {"y": sents[i],
"text": ' '.join(text),
"tag": ' '.join(tag),
"num_words": len(text),
"split": get_split_num(split)}
revs.append(rev_datum)
if all_phrases:
with open(data_file.format("train_phrases"), "rb") as f:
reader = csv.reader(f)
revs_text = []
sents = []
count = 0
for row in reader:
rev, sent = row[0], int(row[1])
rev = clean_str_sst(rev)
if rev in sentence_set:
count += 1
continue
if binary and sent == 2: # skip neutral if binary
continue
rev_tokens = rev.split()
if len(rev_tokens) < min_len:
continue
revs_text.append(rev_tokens)
sent = sentiment_label_for_binary(sent) if binary else sent # check for binary case
sents.append(sent)
revs_tagged = pos_tagger.tag_sents(revs_text)
for i in range(len(revs_tagged)):
rev_tagged = revs_tagged[i]
text = list(zip(*rev_tagged)[0])
tag = list(zip(*rev_tagged)[1])
for word in set(text):
vocab[word] += 1
for postag in set(tag):
pos_vocab[postag] += 1
rev_datum = {"y": sents[i],
"text": ' '.join(text),
"tag": ' '.join(tag),
"num_words": len(text),
"split": get_split_num('train')}
revs.append(rev_datum)
print "{} sentences in phrases".format(count)
return revs, vocab, pos_vocab
def get_W(word_vecs, k):
vocab_size = len(word_vecs)
word_idx_map = dict()
W = np.zeros(shape=(vocab_size+1, k), dtype='float32')
W[0] = np.zeros(k, dtype='float32')
i = 1
for word in word_vecs:
W[i] = word_vecs[word]
word_idx_map[word] = i
i += 1
return W, word_idx_map
def load_bin_vec(fname, vocab):
word_vecs = {}
with open(fname, "rb") as f:
header = f.readline()
vocab_size, layer1_size = map(int, header.split())
binary_len = np.dtype('float32').itemsize * layer1_size
for line in xrange(vocab_size):
word = []
while True:
ch = f.read(1)
if ch == ' ':
word = ''.join(word)
break
if ch != '\n':
word.append(ch)
if word in vocab:
word_vecs[word] = np.fromstring(f.read(binary_len), dtype='float32')
else:
f.read(binary_len)
return word_vecs, layer1_size
def add_unknown_words(word_vecs, vocab, min_df=1, k=300):
for word in vocab:
if word not in word_vecs and vocab[word] >= min_df:
word_vecs[word] = np.random.uniform(-0.25, 0.25, k)
def clean_str_sst(string):
string = re.sub(r"[^A-Za-z0-9(),!?\'\`]", " ", string)
string = re.sub(r"\s{2,}", " ", string)
return string.strip().lower()
if __name__ == "__main__":
w2v_file = "data/GoogleNews-vectors-negative300.bin"
pos_emb_file = "data/1billion-pos-24.bin"
data_file = "sstb/sstb_condensed_{}.csv"
if len(sys.argv) < 3:
arg1, arg2 = 'reviews', 'fine-grained'
else:
arg1, arg2 = sys.argv[1], sys.argv[2]
all_phrases = True if arg1 == 'phrases' else False
binary_case = True if arg2 == 'binary' else False
print "loading sstb data...",
revs, vocab, pos_vocab = build_data_cv(data_file, all_phrases, binary_case)
max_l = np.max(pd.DataFrame(revs)["num_words"])
print "data loaded!"
print "number of sentences: " + str(len(revs))
print "vocab size: " + str(len(vocab))
print "pos vocab size: " + str(len(pos_vocab))
print "max sentence length: " + str(max_l)
print "loading word embeddings...",
w2v, w2v_dim = load_bin_vec(w2v_file, vocab)
print "word embeddings loaded!"
print "pretrained num words: " + str(len(w2v))
add_unknown_words(w2v, vocab, k=w2v_dim)
W, word_idx_map = get_W(w2v, k=w2v_dim)
rand_vecs = {}
add_unknown_words(rand_vecs, vocab, k=w2v_dim)
W_rand, _ = get_W(rand_vecs, k=w2v_dim)
print "loading pos embeddings...",
p2v, p2v_dim = load_bin_vec(pos_emb_file, pos_vocab)
print "pos embeddings loaded!"
print "pretrained num pos tags: " + str(len(p2v))
add_unknown_words(p2v, pos_vocab, k=p2v_dim)
P, pos_idx_map = get_W(p2v, k=p2v_dim)
rand_vecs = {}
add_unknown_words(rand_vecs, pos_vocab, k=p2v_dim)
P_rand, _ = get_W(rand_vecs, k=p2v_dim)
cPickle.dump([revs, W, W_rand, word_idx_map, vocab, P, P_rand, pos_idx_map, 1, 5], open("sstb.p", "wb"))
print "dataset created!"