-
Notifications
You must be signed in to change notification settings - Fork 0
/
elsa_1_run.py
140 lines (110 loc) · 4.78 KB
/
elsa_1_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
"""
This script prepares the running of the different models by doing some
necessary preprocessing, then train-test-splits.
It then creates the template objects for each model and runs every model on
the data.
"""
import datetime
from pathlib import Path
import numpy as np
from sklearn.model_selection import train_test_split
from models.run_transformer import Transformer_template
from models.run_pydts import PyDTS_template
from models.run_deephit import DeepHit_template
from models.run_coxph import CoxPH_template
from utils.utils import Rundir
from utils.elsa_dataset import ELSA_dataset
class ELSA_rundir(Rundir):
"""
Creates a running directory containing the train, val, and test datasets.
renames the columns (this preprocessing is mostly required by the PyDTS
package). Prepares all the parameters that will be useful for initializing
the models.
"""
def __init__(self, basedir, name, random_state):
super().__init__(basedir, name)
self.random_state = random_state
elsa = ELSA_dataset()
self.duration_col = elsa.duration_col
self.event_col = elsa.event_col
self.event_labels = elsa.event_labels
self.label_cols = elsa.label_cols
self.cols = elsa.cols
self.df = elsa.preprocessed_df
self.n_patients = self.df.shape[0]
self.n_cov = self.df.shape[1] - 3
self.params = {
'n_cov': self.n_cov,
'n_times': self.df[self.duration_col].nunique(),
'n_events': self.df[self.event_col].nunique() - 1
}
# Placeholder for ground truth (all nan because unknown)
self.hazard_gt = np.ones((self.n_patients, self.params['n_times']))*np.nan
self.main()
def main(self):
self.savedir, self.plotdir, self.rundir = self.make_rundir()
self.train, test = train_test_split(self.df,
test_size=0.3,
random_state=self.random_state)
self.outcomes = test[[self.event_col, self.duration_col]]
self.test = test.drop(self.label_cols, axis=1)
self.train.to_parquet(f'{self.rundir}/train.parquet')
self.test.to_csv(f'{self.rundir}/test.csv', index=None)
self.outcomes.to_csv(f'{self.rundir}/outcomes.csv')
def _now_str():
return (str(datetime.datetime.now()).split('.')[0]
.replace(':', '_')
.replace('-', '_')
.replace(' ', '_'))
seed = 0
savedir = 'outputs/ELSA/'
rd = ELSA_rundir(basedir=savedir, name=seed, random_state=seed)
weights_savedir = f'models/weights/ELSA_{_now_str()}/'
Path(weights_savedir).mkdir(exist_ok=True, parents=True)
kwargs = ({'train_df': rd.train,
'test_df': rd.test,
'hazard_gt': rd.hazard_gt,
'weights_savedir': weights_savedir}
| rd.params
| rd.cols)
deep_kwargs = {
'batch_size': 128,
'd_model': 64,
'dropout': 0.05
}
coxph_template = CoxPH_template(rd.savedir,
plotdir=f'{rd.plotdir}coxph/',
schoenfeld_savepath='outputs/plots/table_schoenfeld/ELSA_residuals.parquet',
compute_schoenfeld=True,
event_labels=rd.event_labels,
**kwargs)
coxph_template.run()
coxph_template = CoxPH_template(rd.savedir,
plotdir=f'{rd.plotdir}coxph_regularized/',
schoenfeld_savepath='outputs/plots/table_schoenfeld/simu_residuals.parquet',
compute_schoenfeld=False,
name='regularized_CoxPH',
l1_ratio=1,
penalizer=0.01,
**kwargs)
coxph_template.run()
transformer_temp = Transformer_template(rd.savedir,
plotdir=f'{rd.plotdir}transformer/',
patience=10,
use_transformer_decoder=False,
num_heads=1,
lr=1e-4,
**kwargs,
**deep_kwargs)
transformer_temp.run()
deephit_template = DeepHit_template(rd.savedir,
plotdir=f'{rd.plotdir}deephit/',
lr=1e-4,
**kwargs,
**deep_kwargs)
deephit_template.run()
pydts_template = PyDTS_template(rd.savedir,
plotdir=f'{rd.plotdir}pydts/',
savemodel=True,
**kwargs)
pydts_template.run()