-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpreprocess.py
207 lines (192 loc) · 8.1 KB
/
preprocess.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
import nltk
import os
import re
import csv
import random
from tqdm import tqdm
from collections import Counter
from tools import tools_json_dump, tools_get_logger, tools_json_load
def write_data_json(datas, labels, label_map, path):
assert len(datas) == len(labels)
tools_get_logger('preprocess').info(f"write data {len(datas)} {label_map} to {path}")
obj = {
'num': len(datas),
'label_map': label_map,
'data': datas,
'label': labels,
}
tools_json_dump(obj, path)
def clean_text(text:str, lower=True):
text = re.sub(r"[^A-Za-z0-9(),!?\'\`]", " ", text)
text = re.sub(r"\'s", " \'s", text)
text = re.sub(r"\'ve", " \'ve", text)
text = re.sub(r"n\'t", " n\'t", text)
text = re.sub(r"\'re", " \'re", text)
text = re.sub(r"\'d", " \'d", text)
text = re.sub(r"\'ll", " \'ll", text)
text = re.sub(r",", " , ", text)
text = re.sub(r"!", " ! ", text)
text = re.sub(r"\(", " \( ", text)
text = re.sub(r"\)", " \) ", text)
text = re.sub(r"\?", " \? ", text)
text = re.sub(r"\s{2,}", " ", text)
text = ' '.join(nltk.word_tokenize(text)).strip()
return text.lower() if lower else text
def preprocess_imdb(origin_dir, output_dir='./dataset'):
label_map = {
'negative': 0,
'positive': 1
}
for s in ['train', 'test']:
path = f"{origin_dir}/{s}"
path_list = []
dirs = os.listdir(path)
for dir in dirs:
if dir == 'pos' or dir == 'neg':
file_list = os.listdir(os.path.join(path, dir))
file_list = map(lambda x: os.path.join(path, dir, x), file_list)
path_list += list(file_list)
datas = []
labels = []
for p in tqdm(path_list, desc=f'imdb_{s}'):
label = 0 if 'neg' in p else 1
with open(p, 'r', encoding='utf-8') as file:
datas.append(clean_text(file.readline().strip()))
labels.append(label)
write_data_json(datas, labels, label_map, f'{output_dir}/imdb.{s}.json')
def preprocess_sst(path, n_classes, output_dir='./dataset'):
assert n_classes in {2, 5}
judge_sentiment = [0, 0.2, 0.4, 0.6, 0.8, 1.0]
def split_sst():
with open(f"{path}/datasetSentences.txt", 'r') as sentences, open(f"{path}/datasetSplit.txt", 'r') as splits:
train, valid, test = [], [], []
temp = {'1': train, '2': test, '3': valid}
sentences = sentences.readlines()
splits = splits.readlines()
for i in range(1, len(sentences)):
t1 = sentences[i].replace("-LRB-", "(")
t2 = t1.replace("-RRB-", ")")
k = splits[i].strip().split(",")
t = t2.strip().split('\t')
temp[k[1]].append(t[1])
return train, valid, test
def assign_labels(data):
processed_datas = []
processed_labels = []
with open(f"{path}/sentiment_labels.txt", 'r') as labels, open(f"{path}/dictionary.txt", 'r') as tables:
labels = labels.readlines()
tables = tables.readlines()
text2id = {}
for i in range(len(tables)):
s = tables[i].strip().split("|")
text2id[s[0]] = s[1]
id2sentiment = {}
for i in range(len(labels)):
s = labels[i].strip().split("|")
id2sentiment[s[0]] = s[1]
dropped = 0
for line in tqdm(data, desc=f'sst{n_classes}'):
line = line.strip()
if line not in text2id:
dropped += 1
continue
score = float(id2sentiment[text2id[line]])
for i in range(1, len(judge_sentiment)):
if score >= judge_sentiment[i-1] and score <= judge_sentiment[i]:
processed_labels.append(i-1)
break
processed_datas.append(clean_text(line))
assert len(processed_datas) == len(processed_labels)
assert processed_labels[-1] in {0, 1, 2, 3, 4}
if n_classes == 2:
delete = []
for i in range(len(processed_labels)):
if processed_labels[i] < 2: processed_labels[i] = 0
elif processed_labels[i] > 2: processed_labels[i] = 1
else:
delete.append(i)
dropped += len(delete)
for i in sorted(delete, reverse=True):
del processed_datas[i]
del processed_labels[i]
label_map = {'low': 0, 'high': 1}
else:
label_map = {'very low': 0, 'low': 1, 'neutral': 2, 'high': 3, 'very high': 4}
return processed_datas, processed_labels, label_map, dropped
train, valid, test = split_sst()
temp = {'train': train, 'valid': valid, 'test': test}
for k, v in temp.items():
processed_datas, processed_labels, label_map, dropped = assign_labels(v)
tools_get_logger('preprocess').info(f'SST-{n_classes} {k} dropped data {dropped}')
write_data_json(processed_datas, processed_labels, label_map, f'{output_dir}/sst{n_classes}.{k}.json')
def preprocess_agnews(path, output_dir='./dataset'):
types = ['train', 'test']
label_map = {'World': 0, 'Sports': 1, 'Business': 2, 'Sci / Tech': 3}
for t in types:
datas = []
labels = []
with open(f"{path}/{t}.csv", 'r', newline='', encoding='utf-8') as file:
reader = csv.reader(file, delimiter=',')
for line in tqdm(reader, desc=f'agnews_{t}'):
labels.append(int(line[0]) - 1)
line = line[1] + '. ' + line[2]
datas.append(clean_text(line))
write_data_json(datas, labels, label_map, f'{output_dir}/agnews.{t}.json')
def split_data(path, num, output, seed=0):
source_data = tools_json_load(path)
data = source_data['data']
label = source_data['label']
label_map = source_data['label_map']
data_length = source_data['num']
assert num <= data_length
class_num = len(label_map.keys())
keys = list(label_map.values())
num4class = Counter(label)
label4class_list = {}
selected_num4class = {}
for element in keys:
label4class_list[element] = [i for i, x in enumerate(label) if x == element]
count = 0
for i in range(class_num):
if i == class_num - 1:
selected_num4class[keys[i]] = num - count
else:
each_num = int(num * num4class[keys[i]] / data_length)
selected_num4class[keys[i]] = each_num
count += each_num
selected_data = []
selected_label = []
for i in range(class_num):
random.seed(seed + i)
sample_index = random.sample(label4class_list[keys[i]], selected_num4class[keys[i]])
selected_label = selected_label + [label[i] for i in sample_index]
selected_data = selected_data + [data[i] for i in sample_index]
write_data_json(selected_data, selected_label, label_map, output)
if __name__ == '__main__':
import argparse
parser = argparse.ArgumentParser()
choices = ['imdb', 'sst2', 'sst5', 'agnews']
parser.add_argument('--dataset', type=str, choices=choices + ['all'], nargs='+', required=True)
parser.add_argument('--attack_set', type=int, default=1000)
args = parser.parse_args()
if 'all' in args.dataset:
args.dataset = choices
for d in args.dataset:
if d == 'imdb':
preprocess_imdb('./dataset/imdb', output_dir='./dataset')
elif d == 'agnews':
preprocess_agnews('./dataset/ag_news_csv', output_dir='./dataset')
elif d == 'sst2':
preprocess_sst('./dataset/stanfordSentimentTreebank', 2, output_dir='./dataset')
elif d == 'sst5':
preprocess_sst('./dataset/stanfordSentimentTreebank', 5, output_dir='./dataset')
else:
raise NotImplementedError(f"{d} not found")
path = f'./dataset/{d}.test.json'
attack_path = f'./dataset/{d}.attack.json'
if os.path.exists(path):
split_data(path, num=args.attack_set, output=attack_path)
else:
path = f'./dataset/{d}.valid.json'
split_data(path, num=args.attack_set, output=attack_path)
pass