-
Notifications
You must be signed in to change notification settings - Fork 1
/
preprocess.py
236 lines (182 loc) · 8.16 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
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
'''
Created on Aug 1, 2018
@author: Varela
Defines project wide constants
'''
import argparse
import os, glob
import yaml
import pandas as pd
import config
import models.feature_factory as fac
from models.propbank_encoder import PropbankEncoder
from datasets import tfrecords_builder
from utils.info import get_binary
SCHEMA_PATH = '{:}gs.yaml'.format(config.SCHEMA_DIR)
SHIFTS = (-3, -2, -1, 0, 1, 2, 3)
FEATURE_MAKER_DICT = {
'argrecon.csv': {'marker_fnc': lambda x, y: fac.process_argrecon(x, version=y), 'column': 'argument recognition'},
'chunks.csv': {'marker_fnc': lambda x, y: fac.process_chunk(x, version=y), 'column': 'chunk features'},
'ctree_chunk.csv': {'marker_fnc': lambda x, y: fac.process_ctreechunk(x, version=y), 'column': 'shallow chunk features'},
'predicate_marker.csv': {'marker_fnc': lambda x, y: fac.process_predmarker(x, version=y), 'column': 'predicate marker feature'},
'form.csv': {'marker_fnc': lambda x, y: fac.process_shifter_ctx_p(x, ['FORM'], SHIFTS, version=y), 'column': 'form predicate context features'},
'gpos.csv': {'marker_fnc': lambda x, y: fac.process_shifter_ctx_p(x, ['GPOS'], SHIFTS, version=y), 'column': 'gpos predicate context features'},
'lemma.csv': {'marker_fnc': lambda x, y: fac.process_shifter_ctx_p(x, ['LEMMA'], SHIFTS, version=y), 'column': 'lemma predicate context features'},
't.csv': {'marker_fnc': lambda x, y: fac.process_t(x, version=y), 'column': 'chunk label class'},
'iob.csv': {'marker_fnc': lambda x, y: fac.process_iob(x, version=y), 'column': 'iob class'}
}
def make_propbank_encoder(encoder_name='deep_glo50',
language_model='glove_s50', version='1.0', verbose=True):
''' Creates a ProbankEncoder instance from strings.
Arguments:
encoder_name:
language_model:
version
verbose
Returns:
'''
# Process inputs
prefix_dir = config.LANGUAGE_MODEL_DIR
prefix_target_dir = 'datasets/csvs/{:}/'.format(version)
gs_path = '{:}gs.csv'.format(prefix_target_dir)
file_path = '{:}{:}.txt'.format(prefix_dir, language_model)
if not os.path.isfile(file_path):
glob_regex = '{:}*'.format(prefix_dir)
options_list = [
re.sub('\.txt','', re.sub(prefix_dir,'', file_path))
for file_path in glob.glob(glob_regex)]
_errmsg = '{:} not found avalable options are in {:}'
raise ValueError(_errmsg.format(language_model ,options_list))
# Getting to the schema
with open(SCHEMA_PATH, mode='r') as f:
schema_dict = yaml.load(f)
dfgs = pd.read_csv(gs_path, index_col=0, sep=',', encoding='utf-8')
column_files = [
'column_argrecon/argrecon.csv',
'column_chunks/chunks.csv',
'column_ctree_chunks/ctree_chunk.csv',
'column_predmarker/predicate_marker.csv',
'column_shifts_ctx_p/form.csv',
'column_shifts_ctx_p/gpos.csv',
'column_shifts_ctx_p/lemma.csv',
'column_t/t.csv',
'column_iob/iob.csv'
]
gs_dict = dfgs.to_dict()
for column_filename in column_files:
column_path = '{:}{:}'.format(prefix_target_dir, column_filename)
if not os.path.isfile(column_path):
*dirs, filename = column_path.split('/')
# filename = column_path.split('/')[-1]
# dirs = column_path.split('/')[:-1]
dir_ = '/'.join(dirs)
if not os.path.isdir(dir_):
os.makedirs(dir_)
column_dict = FEATURE_MAKER_DICT[filename]
maker_fnc, msg = column_dict['marker_fnc'], column_dict['column']
if verbose:
print('processing:\t{:}'.format(msg))
maker_fnc(gs_dict, version)
column_df = pd.read_csv(column_path, index_col=0, encoding='utf-8')
dfgs = pd.concat((dfgs, column_df), axis=1)
propbank_encoder = PropbankEncoder(
dfgs.to_dict(),
schema_dict,
language_model=language_model,
dbname=encoder_name,
version=version
)
model_, sz_ = language_model.split('_s')
embs_model = '{:}{:}'.format(get_model(model_), sz_)
bin_path = get_binary('deep', embs_model, version=version)
bin_dir = '/'.join(bin_path.split('/')[:-1]) + '/'
if not os.path.isdir(bin_dir):
os.makedirs(bin_dir)
propbank_encoder.persist(bin_dir, filename=encoder_name)
return propbank_encoder
def make_tfrecords(encoder_name='deep_glo50',
propbank_encoder=None,
version='1.0'):
# PREPARE WRITE DIRECTORIES
embs_model = encoder_name.split('_')[-1]
bin_dir = 'datasets/binaries/{:}/'.format(version)
if version in ('1.0'):
bin_dir += '{:}/'.format(embs_model)
if not os.path.isdir(bin_dir):
os.makedirs(bin_dir)
if propbank_encoder is None:
bin_path = '{:}{:}.pickle'.format(bin_dir, encoder_name)
propbank_encoder = PropbankEncoder.recover(bin_path)
cnf_dict = propbank_encoder.to_config(config.DATA_ENCODING)
config.set_config(cnf_dict, embs_model)
flt = None
enc = config.DATA_ENCODING
for ds_type in ('test', 'valid', 'train'):
iter_ = propbank_encoder.iterator(ds_type, filter_columns=flt, encoding=enc)
tfrecords_builder(iter_, ds_type, embs_model, version=version)
def get_model(mname):
if mname == 'wang2vec':
return 'wan'
if mname == 'word2vec':
return 'wrd'
if mname == 'glove':
return 'glo'
return mname[:3]
def get_filename(file_path):
'''Extracts filename from full path
Arguments:
file_path {str} -- sys path
'''
return f.split('/')[-1].replace('.txt', '')
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description='''Preprocess Deep SRL system features and embeddings''')
parser.add_argument('language_model',
type=str, default='glove_s50',
help='''language model for embeddings, more info:
http://nilc.icmc.usp.br/embeddings''')
# This argument now is an environment variable in config module
# parser.add_argument('--encoding', type=str, dest='encoding',
# choices=('HOT', 'EMB', 'MIX', 'IDX'), default='EMB',
# help='''Choice of feature representation based on column type --
# `int`, `bool`, `text`, `choice`. `MIX` will keep `text`
# features as index to be embedded for the input pipeline
# and will one-hot `choice` values. `EMB` will embed `text`
# features and will one-hot encode `choice` features.''')
parser.add_argument('--version', type=str, dest='version',
choices=('1.0', '1.1',), default='1.0',
help='PropBankBr: version 1.0 or 1.1')
args = parser.parse_args()
language_model = args.language_model
version = args.version
print(language_model, version)
lmpath = '{:}{:}.txt'.format(config.LANGUAGE_MODEL_DIR, language_model)
if not os.path.isfile(lmpath):
if not os.path.isdir(config.LANGUAGE_MODEL_DIR):
os.makedirs(config.LANGUAGE_MODEL_DIR)
msg_ = '''{:}:created.
Download cd embeddings
http://nilc.icmc.usp.br/embeddings.'''.\
format(config.LANGUAGE_MODEL_DIR)
raise ValueError(msg_)
else:
glob_regex = '{:}*.txt'.format(config.LANGUAGE_MODEL_DIR)
language_model_list = [
get_filename(f) for f in glob.glob(glob_regex)]
if language_model_list:
raise ValueError('''{:}: not found.
Some avalable options are {:}'''.
format(language_model, language_model_list))
else:
model_, sz_ = language_model.split('_s')
encoder_name = 'deep_{:}{:}'.format(get_model(model_), sz_)
propbank_encoder = make_propbank_encoder(
encoder_name=encoder_name,
language_model=language_model,
version=version
)
make_tfrecords(
encoder_name=encoder_name,
propbank_encoder=propbank_encoder,
version=version,
)