|
a |
|
b/main.py |
|
|
1 |
""" |
|
|
2 |
Main training script. |
|
|
3 |
""" |
|
|
4 |
|
|
|
5 |
import argparse |
|
|
6 |
|
|
|
7 |
from src.utils import training |
|
|
8 |
from src.config import config |
|
|
9 |
|
|
|
10 |
|
|
|
11 |
def main(args): |
|
|
12 |
""" |
|
|
13 |
Runs appropriate experiment(s) from passed args. |
|
|
14 |
""" |
|
|
15 |
if args.models == "all": |
|
|
16 |
args.models = ["MLP", "CNN", "LSTM", "Transformer"] |
|
|
17 |
|
|
|
18 |
if args.strategies == "all": |
|
|
19 |
args.strategies = [ |
|
|
20 |
"Naive", |
|
|
21 |
"Cumulative", |
|
|
22 |
"EWC", |
|
|
23 |
"OnlineEWC", |
|
|
24 |
"SI", |
|
|
25 |
"LwF", |
|
|
26 |
"Replay", |
|
|
27 |
"GEM", |
|
|
28 |
"AGEM", |
|
|
29 |
] |
|
|
30 |
|
|
|
31 |
# Hyperparam optimisation over validation data for first 2 tasks |
|
|
32 |
if args.validate: |
|
|
33 |
training.main( |
|
|
34 |
data=args.data, |
|
|
35 |
domain=args.domain_shift, |
|
|
36 |
outcome=args.outcome, |
|
|
37 |
models=args.models, |
|
|
38 |
strategies=args.strategies, |
|
|
39 |
dropout=args.dropout, |
|
|
40 |
config_generic=config.config_generic, |
|
|
41 |
config_model=config.config_model, |
|
|
42 |
config_cl=config.config_cl, |
|
|
43 |
num_samples=args.num_samples, |
|
|
44 |
validate=True, |
|
|
45 |
) |
|
|
46 |
|
|
|
47 |
# Train and test over all tasks (using optimised hyperparams) |
|
|
48 |
if args.train: |
|
|
49 |
training.main( |
|
|
50 |
data=args.data, |
|
|
51 |
domain=args.domain_shift, |
|
|
52 |
outcome=args.outcome, |
|
|
53 |
models=args.models, |
|
|
54 |
strategies=args.strategies, |
|
|
55 |
) |
|
|
56 |
|
|
|
57 |
|
|
|
58 |
if __name__ == "__main__": |
|
|
59 |
parser = argparse.ArgumentParser() |
|
|
60 |
|
|
|
61 |
parser.add_argument( |
|
|
62 |
"--data", |
|
|
63 |
type=str, |
|
|
64 |
default="mimic3", |
|
|
65 |
choices=["mimic3", "eicu", "random"], |
|
|
66 |
help="Dataset to use.", |
|
|
67 |
) |
|
|
68 |
|
|
|
69 |
parser.add_argument( |
|
|
70 |
"--outcome", |
|
|
71 |
type=str, |
|
|
72 |
default="mortality_48h", |
|
|
73 |
choices=["ARF_4h", "ARF_12h", "Shock_4h", "Shock_12h", "mortality_48h"], |
|
|
74 |
help="Outcome to predict.", |
|
|
75 |
) |
|
|
76 |
|
|
|
77 |
parser.add_argument( |
|
|
78 |
"--domain_shift", |
|
|
79 |
type=str, |
|
|
80 |
default="age", |
|
|
81 |
choices=[ |
|
|
82 |
"time_season", |
|
|
83 |
"region", |
|
|
84 |
"hospital", |
|
|
85 |
"ward", |
|
|
86 |
"age", |
|
|
87 |
"sex", |
|
|
88 |
"ethnicity", |
|
|
89 |
"ethnicity_coarse", |
|
|
90 |
], |
|
|
91 |
help="Domain shift exhibited in tasks.", |
|
|
92 |
) |
|
|
93 |
|
|
|
94 |
parser.add_argument( |
|
|
95 |
"--strategies", |
|
|
96 |
type=str, |
|
|
97 |
default="all", |
|
|
98 |
choices=[ |
|
|
99 |
"Naive", |
|
|
100 |
"Cumulative", |
|
|
101 |
"Joint", |
|
|
102 |
"EWC", |
|
|
103 |
"OnlineEWC", |
|
|
104 |
"SI", |
|
|
105 |
"LwF", |
|
|
106 |
"Replay", |
|
|
107 |
"GDumb", |
|
|
108 |
"GEM", |
|
|
109 |
"AGEM", |
|
|
110 |
], |
|
|
111 |
nargs="+", |
|
|
112 |
help="Continual learning strategy(s) to evaluate.", |
|
|
113 |
) |
|
|
114 |
|
|
|
115 |
parser.add_argument( |
|
|
116 |
"--models", |
|
|
117 |
type=str, |
|
|
118 |
default="all", |
|
|
119 |
choices=["MLP", "CNN", "RNN", "LSTM", "GRU", "Transformer"], |
|
|
120 |
nargs="+", |
|
|
121 |
help="Model(s) to evaluate.", |
|
|
122 |
) |
|
|
123 |
|
|
|
124 |
parser.add_argument( |
|
|
125 |
"--dropout", action="store_true", help="Add dropout to model(s)." |
|
|
126 |
) |
|
|
127 |
|
|
|
128 |
parser.add_argument("--validate", action="store_true", help="Tune hyperparameters.") |
|
|
129 |
|
|
|
130 |
parser.add_argument( |
|
|
131 |
"--train", action="store_true", help="Train and test validated models." |
|
|
132 |
) |
|
|
133 |
|
|
|
134 |
parser.add_argument( |
|
|
135 |
"--num_samples", |
|
|
136 |
type=int, |
|
|
137 |
default=1, |
|
|
138 |
help="Number of samples to draw during hyperparameter search.", |
|
|
139 |
) |
|
|
140 |
|
|
|
141 |
args = parser.parse_args() |
|
|
142 |
main(args) |
