-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathprepro.py
268 lines (220 loc) · 15.5 KB
/
prepro.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
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
# Copyright (c) Microsoft. All rights reserved.
# Modified Copyright by Ubiquitous Knowledge Processing (UKP) Lab, Technische Universität Darmstadt
import argparse
from data_utils.log_wrapper import create_logger
from data_utils.glue_utils import *
import random
LOW_RESOURCE_DATA_RATES = [0.1, 0.3, 0.7]
bert_tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
logger = create_logger(__name__, to_disk=True, log_file='bert_data_proc_512.log')
def get_random_indices(dump_path, data, rate):
rdm_indices = []
if rate < 1.0:
with open("data_splits/low_resource_splits/"+dump_path.rsplit("/", 1)[1].split("_")[0]+"_train.csv", "r") as split_file_in:
rdm_indices = [int(idx) for idx in split_file_in.readline().split(",")[:int(len(data)*rate)]]
return rdm_indices
def build_data(data, dump_path, rate=1.0, MAX_SEQ_LEN=100, is_train=True, tolower=True):
"""Build data of sentence pair tasks
"""
if os.path.isfile(dump_path):
print(dump_path + " already existis. Skipping!")
return
rdm_indices = get_random_indices(dump_path, data, rate)
with open(dump_path, 'w', encoding='utf-8') as writer:
for idx, sample in tqdm(enumerate(data)):
if rate < 1.0 and idx not in rdm_indices:
continue
ids = sample['uid']
premise = bert_tokenizer.tokenize(sample['premise'])
hypothesis = bert_tokenizer.tokenize(sample['hypothesis'])
label = sample['label']
if "spelling" in dump_path or "negation" in dump_path: # have already been truncated in preprocessing step
truncate_len = 512-3
else:
truncate_len = MAX_SEQ_LEN - 3
truncate_seq_pair(premise, hypothesis, truncate_len)
input_ids =bert_tokenizer.convert_tokens_to_ids(['[CLS]'] + hypothesis + ['[SEP]'] + premise + ['[SEP]'])
type_ids = [0] * ( len(hypothesis) + 2) + [1] * (len(premise) + 1)
features = {'uid': ids, 'label': label, 'token_id': input_ids, 'type_id': type_ids, 'hypothesis': sample['hypothesis'], 'premise': sample['premise']}
writer.write('{}\n'.format(json.dumps(features)))
def build_data_single(data, dump_path, rate=1.0, MAX_SEQ_LEN=100):
"""Build data of single sentence tasks
"""
if os.path.isfile(dump_path):
print(dump_path + " already existis. Skipping!")
return
rdm_indices = get_random_indices(dump_path, data, rate)
with open(dump_path, 'w', encoding='utf-8') as writer:
for idx, sample in tqdm(enumerate(data)):
if rate < 1.0 and idx not in rdm_indices:
continue
ids = sample['uid']
premise = bert_tokenizer.tokenize(sample['premise'])
label = sample['label']
if "spelling" in dump_path or "negation" in dump_path: # have already been truncated in preprocessing step
premise = premise[:512 - 3]
elif len(premise) > MAX_SEQ_LEN - 3:
premise = premise[:MAX_SEQ_LEN - 3]
input_ids =bert_tokenizer.convert_tokens_to_ids(['[CLS]'] + premise + ['[SEP]'])
type_ids = [0] * ( len(premise) + 2)
features = {'uid': ids, 'label': label, 'token_id': input_ids, 'type_id': type_ids, 'premise': sample['premise']}
writer.write('{}\n'.format(json.dumps(features)))
def print_status(train_data, dev_data, test_data, test_negation, test_spelling, test_paraphrase, dataset):
logger.info('Loaded {0} {1} train samples'.format(len(train_data), dataset))
logger.info('Loaded {0} {1} dev samples'.format(len(dev_data), dataset))
logger.info('Loaded {0} {1} test samples'.format(len(test_data), dataset))
if test_negation != None:
logger.info('Loaded {0} {1} test negation samples'.format(len(test_negation), dataset))
if test_spelling != None:
logger.info('Loaded {0} {1} test spelling samples'.format(len(test_spelling), dataset))
if test_paraphrase != None:
logger.info('Loaded {0} {1} test paraphras samples'.format(len(test_paraphrase), dataset))
logger.info('\n')
def build_handler(train_data, dev_data, test_data, test_negation, test_spelling, test_paraphrase,
dataset, mt_dnn_root, single=False, MAX_SEQ_LEN=100):
build_fn = build_data_single if single == True else build_data
logger.info('Start converting and storing train/dev/test files for the model')
train_fout = os.path.join(mt_dnn_root, '{0}_train.json'.format(dataset))
dev_fout = os.path.join(mt_dnn_root, '{0}_dev.json'.format(dataset))
test_fout = os.path.join(mt_dnn_root, '{0}_test.json'.format(dataset))
test_negation_fout = os.path.join(mt_dnn_root, '{0}_test_negation.json'.format(dataset))
test_spelling_fout = os.path.join(mt_dnn_root, '{0}_test_spelling.json'.format(dataset))
test_paraphrase_fout = os.path.join(mt_dnn_root, '{0}_test_paraphrase.json'.format(dataset))
for rate in LOW_RESOURCE_DATA_RATES:
build_fn(train_data, train_fout.split(".json")[0]+str(int(rate*100))+"p.json", rate, MAX_SEQ_LEN=MAX_SEQ_LEN)
build_fn(train_data, train_fout)
build_fn(dev_data, dev_fout)
build_fn(test_data, test_fout)
if test_negation != None:
build_fn(test_negation, test_negation_fout)
if test_spelling != None:
build_fn(test_spelling, test_spelling_fout)
if test_paraphrase != None:
build_fn(test_paraphrase, test_paraphrase_fout)
logger.info('done with {0}'.format(dataset))
def parse_args():
parser = argparse.ArgumentParser(description='Preprocessing datasets.')
parser.add_argument('--seed', type=int, default=13)
parser.add_argument('--root_dir', type=str, default='data')
parser.add_argument('--generate_adversarial', type=int, default=0, help="0 if no adversarial samples should be created, otherwise 1")
parser.add_argument('--max_seq_len', type=int, default=100)
parser.add_argument('--opennmt_gpu', type=int, default=-1, help="-1 for CPU, else GPU index.")
parser.add_argument('--opennmt_batch_size', type=int, default=30)
args = parser.parse_args()
return args
def main(args):
root = args.root_dir
seed = args.seed
CREATE_ADVERSARIAL = True if args.generate_adversarial == 1 else False
MAX_SEQ_LEN = args.max_seq_len
OPENNMT_GPU = args.opennmt_gpu # Machine translation on GPU
OPENNMT_BATCH_SIZE = args.opennmt_batch_size # Machine translation model batch size
assert os.path.exists(root)
random.seed(seed)
np.random.seed(seed)
mt_dnn_root = os.path.join(root, 'mt_dnn')
if not os.path.isdir(mt_dnn_root):
os.mkdir(mt_dnn_root)
######################################
# Stance Detection Tasks
######################################
snopes_path = os.path.join(root, 'Snopes/snopes_corpus_2.csv')
argmin_path = os.path.join(root, 'ArgMin/')
fnc1_path = os.path.join(root, 'FNC-1/')
arc_path = os.path.join(root, 'ARC/')
ibm_claim_stance_path = os.path.join(root, 'IBM_CLAIM_STANCE/claim_stance_dataset_v1.csv')
perspectrum_path = os.path.join(root, 'PERSPECTRUM/')
semeval2016t6_path = os.path.join(root, 'SemEval2016Task6/')
semeval2019t7_path = os.path.join(root, 'SemEval2019Task7/')
scd_path = os.path.join(root, 'SCD/')
iac1_path = os.path.join(root, 'IAC/')
######################################
# Loading / Building DATA
######################################
scd_train_data, scd_dev_data, scd_test_data, scd_test_negation, scd_test_spelling, scd_test_paraphrase = \
load_scd(scd_path, GLOBAL_MAP['scd'], create_adversarial=CREATE_ADVERSARIAL, dataset='scd', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU, OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(scd_train_data, scd_dev_data, scd_test_data, scd_test_negation, scd_test_spelling,
scd_test_paraphrase, 'scd')
build_handler(scd_train_data, scd_dev_data, scd_test_data, scd_test_negation,
scd_test_spelling, scd_test_paraphrase, 'scd', mt_dnn_root, single=True, MAX_SEQ_LEN=MAX_SEQ_LEN)
semeval2016t6_train_data, semeval2016t6_dev_data, semeval2016t6_test_data, semeval2016t6_test_negation, \
semeval2016t6_test_spelling, semeval2016t6_test_paraphrase\
= load_semeval2016t6(semeval2016t6_path, GLOBAL_MAP['semeval2016t6'], create_adversarial=CREATE_ADVERSARIAL,
dataset='semeval2016t6', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU, OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(semeval2016t6_train_data, semeval2016t6_dev_data, semeval2016t6_test_data, semeval2016t6_test_negation,
semeval2016t6_test_spelling, semeval2016t6_test_paraphrase, 'semeval2016t6')
build_handler(semeval2016t6_train_data, semeval2016t6_dev_data, semeval2016t6_test_data, semeval2016t6_test_negation,
semeval2016t6_test_spelling, semeval2016t6_test_paraphrase, 'semeval2016t6', mt_dnn_root, MAX_SEQ_LEN=MAX_SEQ_LEN)
ibmcs_train_data, ibmcs_dev_data, ibmcs_test_data, ibmcs_test_negation, ibmcs_test_spelling,\
ibmcs_test_paraphrase = load_ibmcs(ibm_claim_stance_path, GLOBAL_MAP['ibmcs'], create_adversarial=CREATE_ADVERSARIAL,
dataset='ibmcs', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU, OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(ibmcs_train_data, ibmcs_dev_data, ibmcs_test_data, ibmcs_test_negation,
ibmcs_test_spelling, ibmcs_test_paraphrase, 'ibmcs')
build_handler(ibmcs_train_data, ibmcs_dev_data, ibmcs_test_data, ibmcs_test_negation, ibmcs_test_spelling,
ibmcs_test_paraphrase, 'ibmcs', mt_dnn_root, MAX_SEQ_LEN=MAX_SEQ_LEN)
iac1_train_data, iac1_dev_data, iac1_test_data, iac1_test_negation, iac1_test_spelling, iac1_test_paraphrase = \
load_iac1(iac1_path, GLOBAL_MAP['iac1'], create_adversarial=CREATE_ADVERSARIAL, dataset='iac1', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU, OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(iac1_train_data, iac1_dev_data, iac1_test_data, iac1_test_negation, iac1_test_spelling,
iac1_test_paraphrase, 'iac1')
build_handler(iac1_train_data, iac1_dev_data, iac1_test_data, iac1_test_negation, iac1_test_spelling,
iac1_test_paraphrase, 'iac1', mt_dnn_root, MAX_SEQ_LEN=MAX_SEQ_LEN)
snopes_train_data, snopes_dev_data, snopes_test_data, snopes_test_negation,\
snopes_test_spelling, snopes_test_paraphrase = load_snopes(snopes_path, GLOBAL_MAP['snopes'],
create_adversarial=CREATE_ADVERSARIAL,
dataset='snopes', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU, OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(snopes_train_data, snopes_dev_data, snopes_test_data, snopes_test_negation,
snopes_test_spelling, snopes_test_paraphrase, 'snopes')
build_handler(snopes_train_data, snopes_dev_data, snopes_test_data, snopes_test_negation,
snopes_test_spelling, snopes_test_paraphrase, 'snopes', mt_dnn_root, MAX_SEQ_LEN=MAX_SEQ_LEN)
arc_train_data, arc_dev_data, arc_test_data, arc_test_negation, arc_test_spelling, arc_test_paraphrase = \
load_fnc1_arc(arc_path, GLOBAL_MAP['arc'], create_adversarial=CREATE_ADVERSARIAL,
dataset='arc', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU, OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(arc_train_data, arc_dev_data, arc_test_data, arc_test_negation, arc_test_spelling,
arc_test_paraphrase, 'arc')
build_handler(arc_train_data, arc_dev_data, arc_test_data, arc_test_negation, arc_test_spelling,
arc_test_paraphrase, 'arc', mt_dnn_root, MAX_SEQ_LEN=MAX_SEQ_LEN)
semeval2019t7_train_data, semeval2019t7_dev_data, semeval2019t7_test_data, semeval2019t7_test_negation, \
semeval2019t7_test_spelling, semeval2019t7_test_paraphrase\
= load_semeval2019t7(semeval2019t7_path, GLOBAL_MAP['semeval2019t7'], create_adversarial=CREATE_ADVERSARIAL,
dataset='semeval2019t7', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU, OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(semeval2019t7_train_data, semeval2019t7_dev_data, semeval2019t7_test_data, semeval2019t7_test_negation,
semeval2019t7_test_spelling, semeval2019t7_test_paraphrase, 'semeval2019t7')
build_handler(semeval2019t7_train_data, semeval2019t7_dev_data, semeval2019t7_test_data, semeval2019t7_test_negation,
semeval2019t7_test_spelling, semeval2019t7_test_paraphrase, 'semeval2019t7', mt_dnn_root, single=True, MAX_SEQ_LEN=MAX_SEQ_LEN)
perspectrum_train_data, perspectrum_dev_data, perspectrum_test_data, perspectrum_test_negation, \
perspectrum_test_spelling, perspectrum_test_paraphrase = load_perspectrum(perspectrum_path, GLOBAL_MAP['perspectrum'],
create_adversarial=CREATE_ADVERSARIAL,
dataset='perspectrum', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU,
OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(perspectrum_train_data, perspectrum_dev_data, perspectrum_test_data, perspectrum_test_negation,
perspectrum_test_spelling, perspectrum_test_paraphrase, 'perspectrum')
build_handler(perspectrum_train_data, perspectrum_dev_data, perspectrum_test_data, perspectrum_test_negation,
perspectrum_test_spelling, perspectrum_test_paraphrase, 'perspectrum', mt_dnn_root, MAX_SEQ_LEN=MAX_SEQ_LEN)
argmin_train_data, argmin_dev_data, argmin_test_data, argmin_test_negation, \
argmin_test_spelling, argmin_test_paraphrase = load_argmin(argmin_path, GLOBAL_MAP['argmin'],
create_adversarial=CREATE_ADVERSARIAL,
dataset='argmin', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU, OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(argmin_train_data, argmin_dev_data, argmin_test_data, argmin_test_negation,
argmin_test_spelling, argmin_test_paraphrase, 'argmin')
build_handler(argmin_train_data, argmin_dev_data, argmin_test_data, argmin_test_negation,
argmin_test_spelling, argmin_test_paraphrase, 'argmin', mt_dnn_root, MAX_SEQ_LEN=MAX_SEQ_LEN)
fnc1_train_data, fnc1_dev_data, fnc1_test_data, fnc1_test_negation, fnc1_test_spelling,\
fnc1_test_paraphrase = load_fnc1_arc(fnc1_path, GLOBAL_MAP['fnc1'], create_adversarial=CREATE_ADVERSARIAL,
dataset='fnc1', MAX_SEQ_LEN=MAX_SEQ_LEN,
OPENNMT_GPU=OPENNMT_GPU, OPENNMT_BATCH_SIZE=OPENNMT_BATCH_SIZE)
print_status(fnc1_train_data, fnc1_dev_data, fnc1_test_data, fnc1_test_negation,
fnc1_test_spelling, fnc1_test_paraphrase, 'fnc1')
build_handler(fnc1_train_data, fnc1_dev_data, fnc1_test_data, fnc1_test_negation, fnc1_test_spelling,
fnc1_test_paraphrase, 'fnc1', mt_dnn_root, MAX_SEQ_LEN=MAX_SEQ_LEN)
if __name__ == '__main__':
args = parse_args()
main(args)