-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_preparation.py
226 lines (191 loc) · 7.29 KB
/
data_preparation.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
218
219
220
221
222
223
224
225
226
import numpy as np
import pandas as pd
import nltk
import tensorflow_hub as hub
import tensorflow as tf
import pickle
import logging
from tqdm import tqdm
logging.disable(logging.CRITICAL)
elmo_model = hub.Module("https://tfhub.dev/google/elmo/3", trainable=True)
DATA_PATH = './data/'
VUA_PATH = './data/vua/'
TOEFL_PATH = './data/toefl/'
def prepare_vua_data(data_file, save_file):
'''
:param data_file: path to file in data/vua directory from which to read the dataset (csv format)
:param save_file: path to file in data/vua directory to which to write the formatted dataset
'''
data = pd.read_csv(VUA_PATH + data_file)
sentences = []
pos_tags = []
labels = []
for row in data.values:
sen = str(row[2]).split(' ')
sen_len = len(sen)
cleaned_sen = []
labels_array = []
pos_tag = []
for w in sen:
if w.startswith('M_'):
labels_array.append(1)
w1 = w.replace('M_', '')
else:
labels_array.append(0)
w1 = w
cleaned_sen.append(w1)
pos_tag.append(nltk.pos_tag([w1])[0][1])
cleaned_sen = ' '.join(cleaned_sen).strip()
sentences.append(cleaned_sen)
labels.append(labels_array)
pos_tags.append(pos_tag)
try:
assert len(pos_tags) == sen_len
except:
print(cleaned_sen)
print(pos_tags)
# prepare data according to the format of https://github.com/gao-g/metaphor-in-context
df = pd.DataFrame({'txt_id': data['txt_id'].values, 'sen_ix': data['sentence_id'].values,
'sentence': sentences, 'label_seq': labels, 'pos_seq': poss,
'labeled_sentence': data['sentence_txt'].values, 'genre': np.empty_like(labels)})
df.to_csv(VUA_PATH + save_file, index=False)
def prepare_toefl_data(dir_path, save_path):
'''
:param dir_path: path to directory in data/toefl directory which contains the essays
:param save_path: path to file in data/toefl directory to write the formatted dataset
'''
essay_ids = []
sen_ids = []
sentences = []
labeled_sentences = []
label_seq = []
pos_seq = []
for essay_file in tqdm(sorted(os.listdir(TOEFL_PATH + dir_path))):
fp = open(DATA_PATH + dir_path + essay_file, 'r')
essay_id = essay_file[:-4]
i = 1
for line in fp:
sen = str(line)
labels = []
pos_tags = []
cleaned_sen = []
essay_ids.append(essay_id)
sen_ids.append(str(i))
labeled_sentences.append(str(line))
for w in sen.split():
if(w.startswith('M_')):
w1 = w.replace('M_', '')
labels.append(1)
else:
w1 = w
labels.append(0)
pos_tags.append(nltk.pos_tag([w1])[0][1])
cleaned_sen.append(w1)
assert len(labels) == len(sen.split())
assert len(pos_tags) == len(labels)
assert len(cleaned_sen) == len(labels)
sentences.append(' '.join(cleaned_sen))
pos_seq.append(pos_tags)
label_seq.append(labels)
i += 1
df = pd.DataFrame({'txt_id': essay_ids, 'sen_ix': sen_ids, 'sentence': sentences,
'label_seq': label_seq, 'pos_seq': pos_seq,
'labeled_sentence': labeled_sentences, 'genre': list(np.empty_like(label_seq))})
df.to_csv(TOEFL_PATH + save_path, index=False)
def train_val_split(read_path):
'''
:param read_path: path to formatted train dataset file in data/ directory
'''
np.random.seed(101)
df = pd.read_csv(DATA_PATH + read_path)
df = df[df['sentence'].notna()]
# print(len(df))
r = int(len(df)/10)
df = df.sample(frac=1, random_state=1001)
val_df = df[:r]
train_df = df[r:]
val_path = read_path[:-4] + '_val.csv'
train_path = read_path[:-4] + '_train.csv'
val_df.to_csv(DATA_PATH + val_path, index=False)
train_df.to_csv(DATA_PATH + train_path, index=False)
def compute_elmo_vectors(read_path, save_path):
'''
:param read_path: path to formatted dataset files in data/ directory
:param save_path: path to pickle file in data/directory to which to write the elmo vectors
'''
df = pd.read_csv(DATA_PATH + read_path)
dic = {}
txt_ids = df['txt_id'].values
sen_ids = df['sen_ix'].values
sentences = df['sentence'].values
assert len(txt_ids) == len(sentences)
batch_sentences = [sentences[i:min(i+64, len(sentences))] for i in range(0, len(sentences), 64)]
batch_txt_ids = [txt_ids[i:min(i+64, len(txt_ids))] for i in range(0, len(txt_ids), 64)]
batch_sen_ids = [sen_ids[i:min(i+64, len(sen_ids))] for i in range(0, len(sen_ids), 64)]
batch_sen_len = [[len(sen.split(' ')) for sen in batch_sen] for batch_sen in batch_sentences]
assert len(batch_sentences) == len(batch_txt_ids)
for i in tqdm(range(len(batch_sen_ids))):
sen = batch_sentences[i]
txt_id = batch_txt_ids[i]
sen_id = batch_sen_ids[i]
sen_len = batch_sen_len[i]
embed = elmo_model(sen, signature='default', as_dict=True)['elmo']
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
sess.run(tf.tables_initializer())
embeddings = sess.run(embed)
# print(embeddings.shape)
for t, s, e, l in zip(txt_id, sen_id, embeddings, sen_len):
if t not in dic:
dic[t] = {}
dic[t][str(s)] = e[:l, :]
sum = 0
for k in dic:
sum += len(dic[k].keys())
assert len(sentences) == sum
with open(DATA_PATH + save_path, 'wb+') as f:
pickle.dump(dic, f)
def prepare_tokens_data(label_file, save_tokens_file):
'''
:param label_file: path to file in data/ directory which contains the test tokens
:param save_tokens_file: path to pickle file in data/directory to write the offsets
of test tokens corresponding to a sentence
'''
labels = pd.read_csv(DATA_PATH + label_file, header=None)
dic = {}
for row in labels.values:
text_id = row[0].split('_')
txt_id = text_id[0]
sen_id = text_id[1]
offset = int(text_id[2]) - 1
if txt_id not in dic:
dic[txt_id] = {}
if sen_id not in dic[txt_id]:
dic[txt_id][sen_id] = [offset]
else:
dic[txt_id][sen_id].append(offset)
with open(DATA_PATH + save_tokens_file, 'wb+') as f:
pickle.dump(dic, f)
def main():
if sys.argv[1] == 'vua':
prepare_vua_data('vuamc_corpus_train.csv', 'VUA_corpus.csv')
prepare_vua_data('vuamc_corpus_test.csv', 'VUA_corpus_test.csv')
train_val_split('vua/VUA_corpus.csv')
compute_elmo_vectors('vua/VUA_corpus_train.csv', 'vua/elmo_train.pkl')
compute_elmo_vectors('vua/VUA_corpus_val.csv', 'vua/elmo_val.pkl')
compute_elmo_vectors('vua/VUA_corpus_test.csv', 'vua/elmo_test.pkl')
prepare_tokens_data('vua/all_pos_test_tokens.csv', 'vua/all_pos_test_tokens.pkl')
prepare_tokens_data('vua/verb_test_tokens.csv', 'vua/verb_test_tokens.pkl')
elif sys.argv[1] == 'toefl':
prepare_toefl_data('train_essays/', 'TOEFL_corpus.csv')
prepare_toefl_data('test_essays/', 'TOEFL_corpus_test.csv')
train_val_split('toefl/TOEFL_corpus.csv')
compute_elmo_vectors('toefl/TOEFL_corpus_train.csv', 'toefl/elmo_train.pkl')
compute_elmo_vectors('toefl/TOEFL_corpus_val.csv', 'toefl/elmo_val.pkl')
compute_elmo_vectors('toefl/TOEFL_corpus_test.csv', 'toefl/elmo_test.pkl')
prepare_tokens_data('toefl/all_pos_test_tokens.csv', 'toefl/all_pos_test_tokens.pkl')
prepare_tokens_data('toefl/verb_test_tokens.csv', 'toefl/verb_test_tokens.pkl')
else:
print('Unknown option')
if __name__ == "__main__":
main()