forked from XZhang97666/MultimodalMIMIC
-
Notifications
You must be signed in to change notification settings - Fork 0
/
util.py
372 lines (295 loc) · 14.5 KB
/
util.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
import os
import sys
sys.path.insert(0, '../')
sys.path.insert(0, '../TS/mimic3-benchmarks')
sys.path.insert(0, '../ClinicalNotes_TimeSeries/models')
import pickle
import re
import numpy as np
import json
from data import *
import statistics as stat
logger = None
import argparse
import pickle
from accelerate import Accelerator
from sklearn import metrics
from transformers import (AutoTokenizer,
AutoModel,
AutoConfig,
AdamW,
BertTokenizer,
BertModel,
get_scheduler,
set_seed,
BertPreTrainedModel,
LongformerConfig,
LongformerModel,
LongformerTokenizer,
)
def parse_args():
parser = argparse.ArgumentParser(description="Alignment text and ts data")
parser.add_argument(
"--task", type=str, default="ihm"
)
parser.add_argument(
"--file_path", type=str, default="Data", help="A path to dataset folder"
)
parser.add_argument("--output_dir", type=str, default="Checkpoints", help="Where to store the final model.")
parser.add_argument("--tensorboard_dir", type=str, default=None, help="Where to store the final model.")
parser.add_argument("--seed", type=int, default=42, help="A seed for reproducible training.")
parser.add_argument("--mode", type=str, default="train", help="train/test")
parser.add_argument("--modeltype", type=str, default="TS_Text", help="TS, Text or TS_Text")
parser.add_argument("--eval_score", default=['auc', 'auprc', 'f1'], type=list)
parser.add_argument('--num_labels', type=int, default=2)
parser.add_argument("--max_length", type=int, default=128, help=(
"The maximum total input sequence length after tokenization. Sequences longer than this will be truncated," " sequences shorter will be padded if `--pad_to_max_lengh` is passed."),)
parser.add_argument( "--pad_to_max_length", action="store_true", help="If passed, pad all samples to `max_length`. Otherwise, dynamic padding is used.", )
parser.add_argument( "--model_path", type=str, help="Path to pretrained model or model identifier from huggingface.co/models.",
)
parser.add_argument(
"--train_batch_size",
type=int,
default=8,
help="Batch size for the training dataloader.",
)
parser.add_argument(
"--eval_batch_size",
type=int,
default=32,
help="Batch size for the evaluation dataloader.",
)
parser.add_argument("--num_update_bert_epochs", type=int, default=10, help="Number of per training epochs update the bert model.")
parser.add_argument("--num_train_epochs", type=int, default=10, help="Total number of training epochs to perform.")
parser.add_argument(
"--txt_learning_rate",
type=float,
default=5e-5,
help="Initial learning rate for Txt self-attention and Bert to use.",
)
parser.add_argument(
"--ts_learning_rate",
type=float,
default=0.0004,
help="Initial learning rate for TS self-attention to use.",
)
parser.add_argument(
"--gradient_accumulation_steps",
type=int,
default=1,
help="Number of updates steps to accumulate before performing a backward/update pass.",
)
parser.add_argument("--weight_decay", type=float, default=0.01, help="Weight decay to use.")
parser.add_argument(
"--lr_scheduler_type",
type=str,
default="linear",
help="The scheduler type to use.",
choices=["linear", "cosine", "cosine_with_restarts", "polynomial", "constant", "constant_with_warmup"],
)
parser.add_argument( "--pt_mask_ratio",default=0.15, type=float, help="mask rate for pretrain .",
)
parser.add_argument( "--mean_mask_length",default=3, type=int, help="mean mask length for pretrain .",
)
parser.add_argument('--chunk', action='store_true')
parser.add_argument("--chunk_type", default='sent_doc_pos', type=str, help="How to chunk the text. sent_doc_pos: sentence level position + doc level position")
parser.add_argument("--warmup_proportion", default=0.10, type=float, help="proportion for the warmup in the lr scheduler.")
parser.add_argument("--kernel_size", type=int, default=1, help="Kernel size for CNN.")
parser.add_argument("--num_heads", type=int, default=8, help="Number of heads.")
parser.add_argument("--layers", type=int, default=3, help="Number of transformer encoder layer.")
parser.add_argument("--cross_layers", type=int, default=3, help="Number of transformer cross encoder layer.")
parser.add_argument("--embed_dim", default=30, type=int, help="attention embedding dim.")
parser.add_argument("--irregular_learn_emb_ts", action='store_true')
parser.add_argument("--irregular_learn_emb_text", action='store_true')
parser.add_argument("--reg_ts", action='store_true')
parser.add_argument("--tt_max", default=48, type=int, help="max time for irregular time series.")
parser.add_argument("--embed_time", default=64, type=int, help="emdedding for time.")
parser.add_argument('--ts_to_txt', action='store_true')
parser.add_argument('--txt_to_ts', action='store_true')
parser.add_argument("--dropout", default=0.10, type=float, help="dropout.")
parser.add_argument("--model_name", default='BioBert', type=str, help="model for text")
parser.add_argument('--num_of_notes', help='Number of notes to include for a patient input 0 for all the notes', type=int, default=5)
parser.add_argument('--notes_order', help='Should we get notes from beginning of the admission time or from end of it, options are: 1. First: pick first notes 2. Last: pick last notes', default=None)
parser.add_argument('--ratio_notes_order', help='The parameter of a bernulli distribution on whether take notes from First or Last, 1-Last, 0-First',type=float, default=None)
parser.add_argument('--bertcount',type=int, default=3,help='number of count update bert in total')
parser.add_argument('--first_n_item', help='Top n item in val seeds', type=int, default=3)
parser.add_argument('--fine_tune', action='store_true')
parser.add_argument('--self_cross', action='store_true')
parser.add_argument('--TS_mixup', action='store_true', help='mix up reg and irg data')
parser.add_argument("--mixup_level", default=None, type=str, help="mixedup level for two time series data, choose: 'batch', batch_seq' or 'batch_seq_feature'. ")
parser.add_argument('--fp16', action='store_true')
parser.add_argument('--debug', action='store_true')
parser.add_argument('--generate_data', action='store_true')
parser.add_argument('--FTLSTM', action='store_true')
parser.add_argument('--Interp', action='store_true')
parser.add_argument('--cpu', action='store_true')
parser.add_argument("--datagereate_seed", type=int, default=42, help="A seed for reproducible data generation .")
parser.add_argument("--TS_model", type=str, default='Atten', help="LSTM, CNN, Atten")
parser.add_argument("--cross_method", default='self_cross', type=str, help="baseline fusion method: MAGGate, MulT, Outer,concat")
args = parser.parse_args()
return args
def loadBert(args,device):
if args.model_name!=None:
if args.model_name== 'BioBert':
tokenizer = AutoTokenizer.from_pretrained("emilyalsentzer/Bio_ClinicalBERT")
BioBert=AutoModel.from_pretrained("emilyalsentzer/Bio_ClinicalBERT")
elif args.model_name=="bioRoberta":
config = AutoConfig.from_pretrained("allenai/biomed_roberta_base", num_labels=args.num_labels)
tokenizer = AutoTokenizer.from_pretrained("allenai/biomed_roberta_base")
BioBert = AutoModel.from_pretrained("allenai/biomed_roberta_base")
elif args.model_name== "Bert":
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
BioBert = BertModel.from_pretrained("bert-base-uncased")
elif args.model_name== "bioLongformer":
tokenizer = AutoTokenizer.from_pretrained("yikuan8/Clinical-Longformer")
BioBert= AutoModel.from_pretrained("yikuan8/Clinical-Longformer")
else:
raise ValueError("model_name should be BioBert,bioRoberta,bioLongformer or Bert")
else:
if args.model_path!=None:
tokenizer = AutoTokenizer.from_pretrained(args.model_path)
BioBert = AutoModel.from_pretrained(args.model_path)
else:
raise ValueError("provide either model_name or model_path")
BioBert = BioBert.to(device)
BioBertConfig = BioBert.config
return BioBert, BioBertConfig,tokenizer
def data_generate(args):
dataPath = os.path.join(args.file_path, 'all_data_p2x_data.pkl')
if os.path.isfile(dataPath):
print('Using', dataPath)
with open(dataPath, 'rb') as f:
data = pickle.load(f)
if args.debug:
data=data[:100]
data=np.array(data)
total_num=len(data)
idx=np.arange(total_num)
np.random.seed(args.seed)
np.random.shuffle(idx)
train= data[idx[:int(len(idx)*0.8)]]
print(train[0]['data_names'])
val=data[idx[int(len(idx)*0.8):int(len(idx)*0.9)]]
test=data[idx[int(len(idx)*0.9):]]
train=train.tolist()
val=val.tolist()
test=test.tolist()
return train, val, test
def metrics_multilabel(y_true, predictions, verbose=1):
# import pdb; pdb.set_trace()
auc_scores = metrics.roc_auc_score(y_true, predictions, average=None)
ave_auc_micro = metrics.roc_auc_score(y_true, predictions,
average="micro")
ave_auc_macro = metrics.roc_auc_score(y_true, predictions,
average="macro")
ave_auc_weighted = metrics.roc_auc_score(y_true, predictions,
average="weighted")
if verbose:
# print("ROC AUC scores for labels:", auc_scores)
print("ave_auc_micro = {}".format(ave_auc_micro))
print("ave_auc_macro = {}".format(ave_auc_macro))
print("ave_auc_weighted = {}".format(ave_auc_weighted))
return{"auc_scores": auc_scores,
"ave_auc_micro": ave_auc_micro,
"ave_auc_macro": ave_auc_macro,
"ave_auc_weighted": ave_auc_weighted}
def diff_float(time1, time2):
h = (time2-time1).astype('timedelta64[m]').astype(int)
return h/60.0
def get_time_to_end_diffs(times, starttimes):
timetoends = []
for times, st in zip(times, starttimes):
difftimes = []
et = np.datetime64(st) + np.timedelta64(49, 'h')
for t in times:
time = np.datetime64(t)
dt = diff_float(time, et)
assert dt >= 0 #delta t should be positive
difftimes.append(dt)
timetoends.append(difftimes)
return timetoends
def change_data_form(file_path,mode,debug=False):
dataPath = os.path.join(file_path, mode + '.pkl')
if os.path.isfile(dataPath):
# We write the processed data to a pkl file so if we did that already we do not have to pre-process again and this increases the running speed significantly
print('Using', dataPath)
with open(dataPath, 'rb') as f:
# (data, _, _, _) = pickle.load(f)
data = pickle.load(f)
if debug:
data=data[:500]
data_X = data[0]
data_y = data[1]
data_text = data[2]
data_names = data[3]
start_times = data[4]
timetoends = data[5]
dataList=[]
assert len(data_X)==len(data_y)==len(data_text)==len(data_names)==len(start_times)==len(timetoends)
assert len(data_text[0])==len(timetoends[0])
for x,y, text, name, start, end in zip(data_X,data_y,data_text, data_names,start_times,timetoends):
if len(text)==0:
continue
new_text=[]
for t in text:
# import pdb;
# pdb.set_trace()
t=re.sub(r'\s([,;?.!:%"](?:\s|$))', r'\1', t)
t=re.sub(r"\b\s+'\b", r"'", t)
new_text.append(t.lower().strip())
data_detail={"data_names":name,
"TS_data":x,
"text_data":new_text,
"label":y,
"adm_time":start,
"text_time_to_end":end
}
dataList.append(data_detail)
os.makedirs('Data',exist_ok=True)
dataPath2 = os.path.join(file_path, mode + 'p2x_data.pkl')
with open(dataPath2, 'wb') as f:
# Write the processed data to pickle file so it is faster to just read later
pickle.dump(dataList, f)
return dataList
def data_replace(file_path1,file_path2,mode,debug=False):
dataPath1 = os.path.join(file_path2, mode + '.pkl')
dataPath2 = os.path.join(file_path1, mode + 'p2x_data.pkl')
if os.path.isfile(dataPath1):
# We write the processed data to a pkl file so if we did that already we do not have to pre-process again and this increases the running speed significantly
print('Using', dataPath1)
with open(dataPath1, 'rb') as f:
data = pickle.load(f)
if debug:
data=data[:500]
with open(dataPath2, 'rb') as f:
data_r=pickle.load(f)
data_X = data[0]
data_y = data[1]
data_text = data[2]
data_names = data[3]
start_times = data[4]
timetoends = data[5]
data_dict={}
assert len(data_X)==len(data_y)==len(data_text)==len(data_names)==len(start_times)==len(timetoends)
assert len(data_text[0])==len(timetoends[0])
for x,name in zip(data_X, data_names):
data_dict[name]=x
for idx, data_detail in enumerate(data_r):
new_x=data_dict[data_detail['data_names']]
data_detail['TS_data']=new_x
dataPath3=os.path.join(file_path2, mode + 'p2x_data.pkl')
with open(dataPath3, 'wb') as f:
pickle.dump(data_r, f)
def merge_reg_irg(dataPath_reg, dataPath_irg):
with open(dataPath_irg, 'rb') as f:
data_irg=pickle.load(f)
with open(dataPath_reg, 'rb') as f:
data_reg=pickle.load(f)
for idx, data_dict in enumerate(data_reg):
irg_dict=data_irg[data_dict['data_names']]
data_dict['ts_tt']=irg_dict['ts_tt']
data_dict['irg_ts']=irg_dict['irg_ts']
data_dict['irg_ts_mask']=irg_dict['irg_ts_mask']
assert (data_dict['label']==irg_dict['label']).all()
with open(dataPath_reg, 'wb') as f:
pickle.dump(data_reg,f)