-
Notifications
You must be signed in to change notification settings - Fork 3
/
train_val_test_SHINE.py
177 lines (153 loc) · 6.64 KB
/
train_val_test_SHINE.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
import os
import random
import numpy as np
import copy
import torch
import sys
import time
import torch.optim as optim
import pprint as pp
import pandas as pd
import utils.hg_ops as hgo
import pickle as pkl
from SHINE import SHINE
from config import get_config
import getopt
if len(sys.argv) < 2:
sys.exit("Use: python train_val_test_SHINE.py result.csv -c config/config.tcga.SHINE.yaml")
opts, extraparams = getopt.getopt(sys.argv[2:], 's:c:',
['seed=', 'config='])
for o,p in opts:
if o in ['-s', '--seed']:
seed = int(p)
if o in ['-c', '--config']:
fncfg = p
fnres = sys.argv[1]
cfg = get_config(fncfg)
ddn = cfg['data_root']
fn_H = cfg['fn_H']
fn_G = cfg['fn_G']
fn_m = cfg['fn_m']
fn_train = cfg['fn_train']
fn_val = cfg['fn_val']
fn_test = cfg['fn_test']
devstr = cfg['devstr']
dataset = cfg['on_dataset']
rdn = cfg['result_root']
lr_list = cfg['lr']
weight_decay_list = cfg['weight_decay']
gamma_list = cfg['gamma']
n_hid_list = cfg['n_hid']
patience_list=cfg['patience']
atype=cfg['attention_type']
use_subj_edge = cfg['use_subj_edge']
tfidf_H = cfg['tfidf_H']
use_knn = cfg['use_knn']
jk = cfg['jk']
if 'metric' in cfg:
metric = cfg['metric']
else:
metric = 'f1'
if 'seed' in cfg:
seeds = cfg['seed']
else:
seeds = [0]
train_idx = pd.read_csv(fn_train, header=None)[0]
val_idx = pd.read_csv(fn_val, header=None)[0]
test_idx = pd.read_csv(fn_test, header=None)[0]
f = open(fn_m, 'rb')
if dataset == 'MC3':
[m, cf, y] = pkl.load(f)
y, yuniques = pd.factorize(y, sort=True)
cf = None
elif dataset == 'disgenet':
[m, y] = pkl.load(f)
yuniques = y.columns.values
y = y.values
cf = None
else:
sys.exit(f'unrecognized dataset {dataset}')
f.close()
since = time.time()
if use_knn:
H, pathway_idx, knn_idx = hgo.construct_Hexp_KNN_inductive(fn_H, m.iloc[train_idx])
elif use_subj_edge:
H, pathway_idx, subj_idx = hgo.construct_Hexp_inductive(fn_H, m.iloc[train_idx], tfidf = tfidf_H)
else:
H = hgo.construct_H(fn_H, tfidf = tfidf_H)
time_elapsed = time.time() - since
print(f'Constructing H complete in {time_elapsed // 60:.0f}m {time_elapsed % 60:.0f}s')
nfts = np.identity(H.shape[0])
efts = np.identity(H.shape[1])
device = torch.device(devstr)
H = torch.Tensor(np.array(H)).float().to(device)
y = torch.Tensor(y).squeeze().long().to(device)
m = torch.Tensor(np.array(m)).float().to(device)
if cf is None:
dcf = 0
else:
cf = torch.Tensor(np.array(cf)).float().to(device)
dcf = cf.shape[1]
nfts = torch.Tensor(nfts).to(device)
efts = torch.Tensor(efts).to(device)
def _main():
print(f"Classification on {cfg['on_dataset']} dataset!!! classes: {yuniques}")
print('Configuration -> Start')
pp.pprint(cfg)
print('Configuration -> End')
print(f'train_idx.shape: {train_idx.shape}')
fres = open(f'{rdn}/{fnres}', 'w')
fres.write(f'n_hid,lr,weight_decay,gamma,patience,seed,best_train_{metric},best_val_{metric},test_{metric},best_train_loss,best_val_loss,test_loss,best_train_cm,best_val_cm,test_cm\n')
for n_hid in n_hid_list:
for lr in lr_list:
for gamma in gamma_list:
for patience in patience_list:
for weight_decay in weight_decay_list:
for seed in seeds:
fn = f'{rdn}/nhid{n_hid}_lr{lr}_wd{weight_decay}_gamma{gamma}_p{patience}_s{seed}'
if os.path.isfile(f'{fn}.ckpt'):
print(f'Using existing {fn}.ckpt')
else:
np.random.seed(123456789)
model = SHINE(H = H,
yuniques = yuniques,
in_ch_n = nfts.shape[1],
train_idx = train_idx,
val_idx = val_idx,
test_idx = test_idx,
n_hid = n_hid,
jk = jk,
dcf = dcf,
dropout = cfg['drop_out'],
atype = atype,
fn = fn,
metric = metric,
seed = seed,
dataset = dataset)
model = model.to(device)
optimizer = optim.Adam(model.parameters(), lr=lr,
weight_decay=weight_decay)
schedular = optim.lr_scheduler.ReduceLROnPlateau(optimizer,
factor=gamma,
patience=patience)
if dataset == 'MC3':
cls_loss = torch.nn.CrossEntropyLoss(reduction='sum')
elif dataset == 'disgenet':
cls_loss = torch.nn.MultiLabelSoftMarginLoss(reduction='sum')
model = model.fit(nfts, m, cf, y, cls_loss, optimizer, schedular, cfg['max_epoch'], print_freq=cfg['print_freq'])
ckpt = torch.load(f'{fn}.ckpt')
best_train_score = ckpt['best_train_score']
best_train_loss = ckpt['best_train_loss']
best_train_cm = ckpt['best_train_cm']
best_val_score = ckpt['best_val_score']
best_val_loss = ckpt['best_val_loss']
best_val_cm = ckpt['best_val_cm']
test_score = ckpt['test_score']
test_loss = ckpt['test_loss']
test_cm = ckpt['test_cm']
fres.write(f'{n_hid},{lr},{weight_decay},{gamma},{patience},{seed},{best_train_score},{best_val_score},{test_score},{best_train_loss},{best_val_loss},{test_loss},\"{best_train_cm}\",\"{best_val_cm}\",\"{test_cm}\"\n')
print(f'nhid: {n_hid}, lr: {lr}, wd: {weight_decay}, gamma: {gamma}, p: {patience}, s: {seed}')
print(f'Best train {metric}: {best_train_score:4f}, val {metric}: {best_val_score:4f}, test {metric}: {test_score:4f}')
fres.close()
if __name__ == '__main__':
_main()