-
Notifications
You must be signed in to change notification settings - Fork 10
/
run.py
243 lines (207 loc) · 9.72 KB
/
run.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
import argparse
import logging
import shutil
from pathlib import Path
import pandas as pd
import numpy as np
import torch
from tqdm import tqdm
from attrdict import AttrDict
from models.model_pipelines import Model
from models.gap.exec import fit_fold
from models.utils import init_coref_models, init_data
tqdm.pandas(desc="Applying..")
logger= logging.getLogger("GAP")
syslog = logging.StreamHandler()
formatter = logging.Formatter('%(message)s')
syslog.setFormatter(formatter)
logger.setLevel(logging.INFO)
logger.handlers = []
logger.addHandler(syslog)
logger.propagate = False
def run(verbose=0,
model_version=None,
coref_models=[],
data_dir=None,
exp_dir=None,
do_preprocess_train=False,
do_preprocess_eval=False,
force=False,
**kwargs):
args = AttrDict(kwargs)
exp_dir = Path(exp_dir)
logging.getLogger('steppy').setLevel(logging.INFO)
if verbose == 0:
logging.getLogger('steppy').setLevel(logging.WARNING)
if do_preprocess_train or do_preprocess_eval:
if do_preprocess_train and force:
shutil.rmtree(exp_dir / 'data_pipeline', ignore_errors=True)
if do_preprocess_eval and force:
# remove eval data
shutil.rmtree(exp_dir / 'data_pipeline' / 'test', ignore_errors=True)
if model_version == 'grep':
coref_models_ = init_coref_models(coref_models)
else:
coref_models_ = []
else:
coref_models_ = {name: None for name in coref_models}
annotate_coref_mentions = pretrained_proref = model_version == 'grep'
X_trn, X_val, X_tst, X_neither, X_inference = init_data(data_dir,
exp_dir,
persist=True,
sanitize_labels=args.sanitize_labels,
annotate_coref_mentions=annotate_coref_mentions,
pretrained_proref=pretrained_proref,
coref_models=coref_models_,
test_path=args.test_path,
verbose=verbose)
if args.do_train or args.do_eval:
n_gpu = torch.cuda.device_count()
n_samples = 0
if n_gpu == 4:
n_samples = 3
if n_gpu == 8:
n_samples = 8
if args.do_kaggle:
res = Model().ensembled_lms(fit_fold,
pd.concat([X_trn,
X_val,
X_tst,
X_neither,
X_neither.head(n_samples)]).reset_index(drop=True),
None,
X_tst=X_inference,
seeds=args.seeds,
n_folds=args.n_folds,
lms=args.lms,
exp_dir=exp_dir,
sub_sample_path=args.sub_sample_path,
verbose=verbose,
parameters = {
'do_train': args.do_train,
'do_eval': args.do_eval,
'max_seq_length': args.max_seq_length,
'train_batch_size': args.train_batch_size,
'eval_batch_size': args.eval_batch_size,
'learning_rate': args.learning_rate,
'num_train_epochs': args.num_train_epochs,
'patience': args.patience,
'model_version': model_version,
'n_coref_models': len(coref_models)
}
)
else:
if args.test_path:
X_tst = X_inference
res = Model().train_evaluate(fit_fold,
X_trn,
X_val,
X_tst=X_tst,
seed=args.seeds[0],
lm=args.lms[0],
exp_dir=exp_dir,
sub_sample_path=args.sub_sample_path,
test_path=args.test_path,
verbose=verbose,
parameters = {
'do_train': args.do_train,
'do_eval': args.do_eval,
'max_seq_length': args.max_seq_length,
'train_batch_size': args.train_batch_size,
'eval_batch_size': args.eval_batch_size,
'learning_rate': args.learning_rate,
'num_train_epochs': args.num_train_epochs,
'patience': args.patience,
'model_version': model_version,
'n_coref_models': len(coref_models)
}
)
return res
def main():
import os
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"]="0,1,2"
parser = argparse.ArgumentParser()
## Required parameters
parser.add_argument("--model",
default='grep',
type=str,
choices=['probert', 'grep'],
help="probert or grep")
parser.add_argument("--language_model",
default='bert-base-uncased',
type=str,
choices=['bert-base-uncased', 'bert-large-uncased', 'bert-base-cased', 'bert-large-cased'],
help="Lanugage model to be used. In Kaggle mode, the predictions will be averaged over all runs.")
parser.add_argument("--coref_models",
default='url,allen,hug,lee',
type=str,
help="Coref models to be used by GREP. Syntactic distance, Parallelism, Parallelism+URL, \
AllenNLP, Huggingface NeuralCoref, e2e coref by Lee Et Al. Choices are 'syn', 'par', 'url', 'allen', 'hug', 'lee'")
parser.add_argument("--preprocess_train",
default=False,
action='store_true')
parser.add_argument("--preprocess_eval",
default=False,
action='store_true')
parser.add_argument("--train",
default=False,
action='store_true',
help="Whether to run training.")
parser.add_argument("--predict",
default=False,
action='store_true',
help="Whether to predict on the test set.")
parser.add_argument("--kaggle",
default=False,
action='store_true',
help="If true all of the data will be used for training. Otherwise, only gap-development will be used.")
parser.add_argument("--data_dir",
default='data/',
type=str)
parser.add_argument("--exp_dir",
required=True,
type=str,
help="The output directory where the model checkpoints will be written.")
parser.add_argument("--test_path",
default=None,
type=str)
parser.add_argument("--sub_sample_path",
default=None,
type=str)
parser.add_argument("--verbose",
default=0,
type=int)
parser.add_argument("--force",
default=False,
action='store_true',
help='Force clears all cached data.')
args = parser.parse_args()
lms = args.language_model.split(',')
coref_models = args.coref_models.split(',')
res = run(model_version=args.model,
lms=lms,
coref_models=coref_models,
sanitize_labels=True,
seeds=[42, 59, 75, 46, 91],
n_folds=5,
do_preprocess_train=args.preprocess_train,
do_preprocess_eval=args.preprocess_eval,
do_train=args.train,
do_eval=args.predict,
do_kaggle=args.kaggle,
data_dir=args.data_dir,
exp_dir=args.exp_dir,
test_path=args.test_path,
sub_sample_path=args.sub_sample_path,
max_seq_length=512,
train_batch_size=6,
eval_batch_size=32,
learning_rate=4e-6,
num_train_epochs=20,
patience=3,
verbose=args.verbose,
force=args.force
)
if __name__ == "__main__":
main()