-
Notifications
You must be signed in to change notification settings - Fork 0
/
run.py
439 lines (366 loc) · 21.6 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
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
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
# -*- coding: utf-8 -*-
import argparse
import functools
import logging
import os
import re
import sys
from pathlib import Path
from typing import Sequence, Optional
import pandas as pd
from icu_benchmarks.data.feature_extraction import extract_feature_df
from icu_benchmarks.common.datasets import Dataset
from icu_benchmarks.common.lookups import read_var_ref_table
from icu_benchmarks.common.processing import map_df
from icu_benchmarks.common.reference_data import read_static
from icu_benchmarks.common.resampling import irregular_to_gridded
from icu_benchmarks.common.constants import MORTALITY_NAME, CIRC_FAILURE_NAME, RESP_FAILURE_NAME, URINE_REG_NAME, \
URINE_BINARY_NAME, PHENOTYPING_NAME, LOS_NAME
from icu_benchmarks.data import imputation_for_endpoints, extended_general_table_generation, endpoint_generation, \
labels, schemata
from icu_benchmarks.data.preprocess import to_ml
from icu_benchmarks.models.train import train_with_gin
from icu_benchmarks.models.utils import get_bindings_and_params
from icu_benchmarks.preprocessing import merge
default_seed = 42
def build_parser():
parser = argparse.ArgumentParser(
description='Benchmark lib for processing and evaluation of deep learning models on HiRID ICU data')
parent_parser = argparse.ArgumentParser(add_help=False)
subparsers = parser.add_subparsers(title='Commands',
dest='command', required=True)
parser_prep_ml = subparsers.add_parser('preprocess',
help='Calls sequentially merge and resample.',
parents=[parent_parser])
preprocess_arguments = parser_prep_ml.add_argument_group(
'Preprocess arguments')
preprocess_arguments.add_argument('--work-dir',
required=False, type=Path,
help="")
preprocess_arguments.add_argument('--hirid-data-root',
type=Path,
required=True,
help="Path to the decompressed parquet data directory as published on physionet.")
preprocess_arguments.add_argument('--var-ref-path', dest="var_ref_path",
required=True, type=Path,
help="Path to load the variable references from ")
preprocess_arguments.add_argument('-nw', '--nr-workers', default=1,
required=False, type=int,
dest='nr_workers',
help='Number of process to use at preprocessing, Default to 1 ')
preprocess_arguments.add_argument('--split-path', dest="split_path",
default=None, required=False, type=Path,
help="Path to load the data split from from ")
preprocess_arguments.add_argument('--seed', dest="seed",
default=default_seed, required=False, type=int,
help="Seed for the train/val/test split")
preprocess_arguments.add_argument('--imputation', dest="imputation",
default='ffill', required=False, type=str,
help="Type of imputation. Default: 'ffill' ")
preprocess_arguments.add_argument('--horizon', dest="horizon",
default=12, required=False, type=int,
help="Horizon of prediction in hours for failure tasks")
model_arguments = parent_parser.add_argument_group('Model arguments')
model_arguments.add_argument('-l', '--logdir', dest="logdir",
required=False, type=str,
help="Path to the log directory ")
model_arguments.add_argument('--reproducible', default=True, dest="reproducible",
required=False, type=str,
help="Whether to configure torch to be reproducible.")
model_arguments.add_argument('-sd', '--seed', default=1111, dest="seed",
required=False, nargs='+', type=int,
help="Random seed at training and evaluation, default : 1111")
model_arguments.add_argument('-t', '--task', default=None, dest="task",
required=False, nargs='+', type=str,
help="Name of the task : Default None")
model_arguments.add_argument('-r', '--resampling', default=None, dest="res",
required=False, type=int,
help="resampling for the data")
model_arguments.add_argument('-rl', '--resampling_label', default=None,
dest="res_lab", required=False, type=int,
help="resampling for the prediction")
model_arguments.add_argument('-bs', '--batch-size', default=None,
dest="batch_size", required=False,
type=int, nargs='+',
help="Batchsize for the model")
model_arguments.add_argument('-lr', '--learning-rate', default=None, nargs='+',
dest="lr", required=False, type=float,
help="Learning rate for the model")
model_arguments.add_argument('--maxlen', default=None, dest="maxlen",
required=False, type=int,
help="Max length of considered time-series for the model")
model_arguments.add_argument('--num-class', default=None, dest="num_class",
required=False, type=int,
help="Number of classes considered for the task")
model_arguments.add_argument('-emb', '--emb', default=None, dest="emb",
required=False, nargs='+', type=int,
help="Embedding size of the input data")
model_arguments.add_argument('-kernel', '--kernel', default=None,
dest="kernel", required=False, nargs='+',
type=int, help="Kernel size for Temporal CNN")
model_arguments.add_argument('-do', '--do', default=None, dest="do",
required=False, nargs='+', type=float,
help="Dropout probability for the Transformer block")
model_arguments.add_argument('-do_att', '--do_att', default=None, dest="do_att",
required=False, nargs='+', type=float,
help="Dropout probability for the Self-Attention layer only")
model_arguments.add_argument('-depth', '--depth', default=None,
dest="depth", required=False, nargs='+',
type=int,
help="Number of layers in Neyral Network")
model_arguments.add_argument('-heads', '--heads', default=None,
dest="heads", required=False, nargs='+',
type=int,
help="Number of heads in Sel-Attention layer")
model_arguments.add_argument('-latent', '--latent', default=None,
dest="latent", required=False, nargs='+',
type=int,
help="Dimension of fully-conected layer in Transformer block")
model_arguments.add_argument('-horizon', '--horizon', default=None,
dest="horizon", required=False, nargs='+',
type=int,
help="History length for Neural Networks")
model_arguments.add_argument('-hidden', '--hidden', default=None,
dest="hidden", required=False, nargs='+',
type=int,
help="Dimensionality of hidden layer in Neural Networks")
model_arguments.add_argument('--subsample-data', default=None,
dest="subsample_data", required=False, nargs='+',
type=float,
help="Subsample parameter in Gradient Boosting, subsample ratio of the training instance")
model_arguments.add_argument('--subsample-feat', default=None,
dest="subsample_feat", required=False, nargs='+',
type=float,
help="Colsample_bytree parameter in Gradient Boosting, subsample ratio of columns when constructing each tree")
model_arguments.add_argument('--regularization', default=None,
dest="regularization", required=False, nargs='+',
type=float,
help="L1 or L2 regularization type")
model_arguments.add_argument('-rs', '--random-search', default=False,
dest="rs", required=False, type=bool,
help="Random Search setting")
model_arguments.add_argument('-c_parameter', '--c_parameter', default=None,
dest="c_parameter", required=False, nargs='+',
help="C parameter in Logistic Regression")
model_arguments.add_argument('-penalty', '--penalty', default=None,
dest="penalty", required=False, nargs='+',
help="Penalty parameter for Logistic Regression")
model_arguments.add_argument('--loss-weight', default=None,
dest="loss_weight", required=False, nargs='+', type=str,
help="Loss weigthing parameter")
model_arguments.add_argument('-o', '--overwrite', default=False,
dest="overwrite", required=False, type=bool,
help="Boolean to overwrite previous model in logdir")
model_arguments.add_argument('-c', '--config', default=None, dest="config",
nargs='+', type=str,
help="Path to the gin train config file.")
model_arguments.add_argument('-pool', '--pool', default=None, dest="pool",
required=False, nargs='+', type=str, help="pooling method used in RTDLTransformer")
model_arguments.add_argument('-proj', '--proj', action='store_true', dest="proj")
model_arguments.add_argument('-freeze', '--freeze', action='store_true', dest="freeze")
model_arguments.add_argument('-gamma', '--gamma', default=None, required=False, nargs='+', dest="gamma", type=float)
model_arguments.add_argument('-add_cls', '--add_cls', action='store_true', dest="add_cls")
model_arguments.add_argument('-cls_only', '--cls_only', action='store_true', dest="cls_only")
model_arguments.add_argument('-rtdl_emb', '--rtdl_emb', type=int, default=None, dest="rtdl_emb")
model_arguments.add_argument('-path_to_ckpt', '--path_to_ckpt', type=str, default=None, dest="path_to_ckpt")
model_arguments.add_argument('-frac', '--frac', type=float, default=1.0, dest="frac")
parser_evaluate = subparsers.add_parser('evaluate', help='evaluate',
parents=[parent_parser])
parser_train = subparsers.add_parser('train', help='train',
parents=[parent_parser])
return parser
def run_merge_step(hirid_path, var_ref_path, merged_path, nr_workers, static_data_path=None, part_nr=None):
static_data_path = Path(static_data_path) if static_data_path else hirid_path / 'general_table'
if not Dataset(merged_path).is_done():
logging.info("Running merge step...")
merge.merge_tables(
hirid_path / 'observation_tables' / 'parquet' if not part_nr else hirid_path / 'observation_tables' / 'parquet' / f'part-{part_nr}.parquet',
hirid_path / 'pharma_records' / 'parquet' if not part_nr else hirid_path / 'pharma_records' / 'parquet' / f'part-{part_nr}.parquet',
static_data_path,
var_ref_path,
merged_path,
nr_workers
)
else:
logging.info(f"Skipping merging, as outputdata seems to exist in {merged_path}")
def run_resample_step(merged_path: Path, static_path, var_ref_path, common_path, nr_workers: int):
merged_ds = Dataset(merged_path)
output_ds = Dataset(common_path)
if output_ds.is_done():
logging.info(f"Skipping resampling, as outputdata seems to exist in {common_path}")
return
logging.info("Running resample step...")
parts = merged_ds.list_parts()
df_static = read_static(static_path)
df_var_ref = read_var_ref_table(var_ref_path)
prepare_data_fn = functools.partial(irregular_to_gridded,
df_static=df_static,
df_var_ref=df_var_ref)
output_ds.prepare()
map_df(prepare_data_fn, parts,
lambda p: pd.read_parquet(p),
lambda df, p: df.to_parquet(common_path / p,
index=False), nr_workers)
output_ds.mark_done()
def run_feature_extraction_step(common_path: Path, var_ref_path, feature_path, nr_workers: int):
common_ds = Dataset(common_path)
feature_ds = Dataset(feature_path)
if feature_ds.is_done():
logging.info(f"Skipping feature extraction, as outputdata seems to exist in {feature_path}")
return
logging.info("Running feature extraction step")
parts = common_ds.list_parts()
df_var_ref = read_var_ref_table(var_ref_path)
prepare_data_fn = functools.partial(extract_feature_df,
df_var_ref=df_var_ref)
feature_ds.prepare()
map_df(prepare_data_fn, parts,
lambda p: pd.read_parquet(p),
lambda df, p: df.to_parquet(feature_path / p,
index=False), nr_workers)
feature_ds.mark_done()
def run_build_ml(common_path, labels_path, features_path: Optional[Path], ml_path, var_ref_path,
endpoint_names: Sequence[str],
imputation: str, seed: int, split_path=None):
common_ds = Dataset(common_path)
parts = common_ds.list_parts()
labels_ds = Dataset(labels_path, part_re=re.compile('batch_([0-9]+).parquet'))
labels = labels_ds.list_parts()
if features_path:
features_ds = Dataset(features_path)
features = features_ds.list_parts()
else:
features = []
df_var_ref = read_var_ref_table(var_ref_path)
output_ds = Dataset(ml_path)
output_cols = schemata.cols_ml_stage_v1
if not output_ds.is_done():
logging.info("Running build_ml")
output_ds.prepare(single_part=True)
to_ml(ml_path, parts, labels, features, endpoint_names, df_var_ref,
imputation, output_cols, split_path=split_path,
random_seed=seed)
else:
logging.info(f"Data in {ml_path} seem to exist, skipping")
def _get_general_data_path(general_data_path, hirid_data_root):
if general_data_path:
return Path(general_data_path)
else:
general_data_path_physionet_download = (hirid_data_root / 'general_table.csv')
if general_data_path_physionet_download.exists():
return general_data_path_physionet_download
return hirid_data_root / 'general_table'
def run_preprocessing_pipeline(hirid_data_root, work_dir, var_ref_path, imputation_method,
general_data_path=None, split_path=None, seed=default_seed, nr_workers=1, horizon=12):
work_dir.mkdir(exist_ok=True, parents=True)
general_data_path = _get_general_data_path(general_data_path, hirid_data_root)
extended_general_data_path = work_dir / 'general_table_extended.parquet'
if not extended_general_data_path.exists():
logging.info(f"Generating extended general table in {extended_general_data_path}")
extended_general_table_generation.generate_extended_general_table(hirid_data_root / 'observation_tables' / 'parquet',
general_data_path,
extended_general_data_path)
else:
logging.info(f"Using extended general table in {extended_general_data_path}")
merged_path = work_dir / 'merged_stage'
imputation_for_endpoints_path = work_dir / 'imputation_for_endpoints'
endpoints_path = work_dir / 'endpoints'
common_path = work_dir / 'common_stage'
label_name = "_".join(['labels', str(horizon)]) + 'h'
label_path = work_dir / label_name
features_path = work_dir / 'features_stage'
ml_name = 'ml_stage' + '_' + str(horizon) + 'h' + '.h5'
ml_path = work_dir / 'ml_stage' / ml_name
run_merge_step(hirid_data_root, var_ref_path, merged_path, nr_workers, extended_general_data_path)
run_resample_step(merged_path, extended_general_data_path, var_ref_path, common_path, nr_workers)
if not imputation_for_endpoints_path.exists():
logging.info("Running imputation step for endpoints")
imputation_for_endpoints.impute_for_endpoints(merged_path, imputation_for_endpoints_path, nr_workers=nr_workers)
else:
logging.info(f"Data for imputation for endpoints in {imputation_for_endpoints_path} seems to exist, skipping")
if not endpoints_path.exists():
logging.info("Running endpoint generation")
endpoint_generation.generate_endpoints(merged_path, imputation_for_endpoints_path, endpoints_path,
nr_workers=nr_workers)
else:
logging.info(f"Endpoints in {endpoints_path} seem to exist, skipping")
if not label_path.exists():
logging.info("Running label generation")
labels.generate_labels(endpoints_path, imputation_for_endpoints_path, extended_general_data_path, label_path,
nr_workers=nr_workers)
else:
logging.info(f"Labels in {label_path} seem to exist, skipping")
run_feature_extraction_step(common_path, var_ref_path, features_path, nr_workers)
endpoints = (MORTALITY_NAME,
CIRC_FAILURE_NAME + '_' + str(horizon) + 'Hours',
RESP_FAILURE_NAME + '_' + str(horizon) + 'Hours',
URINE_REG_NAME,
URINE_BINARY_NAME,
PHENOTYPING_NAME,
LOS_NAME)
run_build_ml(common_path, label_path, features_path, ml_path, var_ref_path, endpoints,
imputation_method, seed, split_path)
def main(my_args=tuple(sys.argv[1:])):
args = build_parser().parse_args(my_args)
log_fmt = '%(asctime)s - %(levelname)s: %(message)s'
logging.basicConfig(format=log_fmt)
logging.getLogger().setLevel(logging.INFO)
# Dispatch
if args.command == 'preprocess':
run_preprocessing_pipeline(args.hirid_data_root, args.work_dir, args.var_ref_path,
imputation_method=args.imputation,
split_path=args.split_path,
seed=args.seed, nr_workers=args.nr_workers, horizon=args.horizon)
if args.command in ['train', 'evaluate']:
load_weights = args.command == 'evaluate'
reproducible = str(args.reproducible) == 'True'
if not isinstance(args.seed, list):
seeds = [args.seed]
else:
seeds = args.seed
if not load_weights:
gin_bindings, log_dir = get_bindings_and_params(args)
else:
gin_bindings, _ = get_bindings_and_params(args)
log_dir = args.logdir
if args.rs:
reproducible = False
max_attempt = 0
is_already_ran = os.path.isdir(log_dir)
while is_already_ran and max_attempt < 500:
gin_bindings, log_dir = get_bindings_and_params(args)
is_already_ran = os.path.isdir(log_dir)
max_attempt += 1
if max_attempt >= 300:
raise Exception('Reached max attempt to find unexplored set of parameters parameters')
if args.task is not None:
for task in args.task:
gin_bindings_task = gin_bindings + [
'TASK = ' + "'" + str(task) + "'"]
log_dir_task = os.path.join(log_dir, str(task))
for seed in seeds:
if not load_weights:
log_dir_seed = os.path.join(log_dir_task, str(seed))
else:
log_dir_seed = log_dir_task
train_with_gin(model_dir=log_dir_seed,
overwrite=args.overwrite,
load_weights=load_weights,
gin_config_files=args.config,
gin_bindings=gin_bindings_task,
seed=seed, reproducible=reproducible)
else:
for seed in seeds:
if not load_weights:
log_dir_seed = os.path.join(log_dir, str(seed))
else:
log_dir_seed = log_dir
train_with_gin(model_dir=log_dir_seed,
overwrite=args.overwrite,
load_weights=load_weights,
gin_config_files=args.config,
gin_bindings=gin_bindings,
seed=seed, reproducible=reproducible)
"""Main module."""
if __name__ == '__main__':
main()