forked from xtianmcd/GCNeuro
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
319 lines (293 loc) · 11.7 KB
/
train.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
import numpy as np
from scipy.stats import chisquare as X2
import torch.nn.functional as F
from NN import *
from GCN import *
from utils import *
import math
import sys
def train(epoch, adj, all_feats, labels, model, optimizer,n):
t = time.time()
model.train()
o=[]
l=[]
optimizer.zero_grad()
for f in range(len(all_feats)):
o.append(model(all_feats[f], adj,n))
# #if f+1%100==0:
# # output=torch.stack(o)
# # loss_train = F.nll_loss(output,labels[f-99:f+1])
# # acc_train = accuracy(output, labels)
# # loss_train.backward()
# # optimizer.step()
# # o=[]
# loss_train = F.nll_loss(op, labels[f])
# loss_train.backward()
# optimizer.step()
output=torch.stack(o)
loss_train = F.nll_loss(output, labels)
acc_train = accuracy(output, labels)
loss_train.backward()
optimizer.step()
# model.eval()
# loss_val = F.nll_loss(output[idx_val], labels[idx_val])
# acc_val = accuracy(output[idx_val], labels[idx_val])
print('Epoch: {:04d}'.format(epoch+1),
'loss_train: {:.4f}'.format(loss_train.item()),
'acc_train: {:.4f}'.format(acc_train.item()),
'time: {:.4f}s'.format(time.time() - t))
#'loss_val: {:.4f}'.format(loss_val.item()),
#'acc_val: {:.4f}'.format(acc_val.item()),
#'time: {:.4f}s'.format(time.time() - t))
return loss_train, acc_train #loss_val, acc_val, loss_train, acc_train
def test(adj, features, labels,model,n):
model.eval()
with torch.no_grad():
o=[]
for feats in features:
output = model(feats, adj,n)
o.append(output)
output=torch.stack(o)
loss_test = F.nll_loss(output, labels)
acc_test = accuracy(output, labels)
o_pos = [out[1] for out in output.detach()]
auc = sklearn.metrics.roc_auc_score(labels,o_pos)
print("Test set results:",
"loss= {:.4f}".format(loss_test.item()),
"accuracy= {:.4f}".format(acc_test.item()))
return output, loss_test, acc_test, auc
if __name__=='__main__':
# Training settings
# parser = argparse.ArgumentParser()
# parser.add_argument('--no-cuda', action='store_true', default=False,
# help='Disables CUDA training.')
# parser.add_argument('--fastmode', action='store_true', default=False,
# help='Validate during training pass.')
# parser.add_argument('--seed', type=int, default=42, help='Random seed.')
# parser.add_argument('--epochs', type=int, default=200,
# help='Number of epochs to train.')
# parser.add_argument('--lr', type=float, default=0.01,
# help='Initial learning rate.')
# parser.add_argument('--weight_decay', type=float, default=5e-4,
# help='Weight decay (L2 loss on parameters).')
# parser.add_argument('--hidden', type=int, default=16,
# help='Number of hidden units.')
# parser.add_argument('--dropout', type=float, default=0.5,
# help='Dropout rate (1 - keep probability).')
no_cuda=True
# args = parser.parse_args()
# args.cuda = not args.no_cuda and torch.cuda.is_available()
use_cuda = not no_cuda and torch.cuda.is_available()
main_dir = sys.argv[1]
adj, pt_features, pt_labels, skl_features, skl_labels, pt_te_features,
pt_te_labels, skl_te_features, skl_te_labels = load_data(main_dir)
print('SUMMARY')
print(f'adj matrix shape: {adj.shape}')
print(f'skl features[0][0]: {len(skl_features[0][0])}')
print(f'skl features[0]: {len(skl_features[0])}')
print(f'skl features: {len(skl_features)}')
print(f'skl labels[0]: {skl_labels[0].shape}')
print(f'skl labels: {len(skl_labels)}')
print(f'pt features[0][0]: {len(pt_features[0][0])}')
print(f'pt features[0] {len(pt_features[0])}')
print(f'pt features: {len(pt_features)}')
print(f'pt labels[0]: {pt_labels[0].shape}')
print(f'pt labels: {len(pt_labels)}')
bl_te_acc={}
bl_te_f1={}
bl_te_auc={}
seed=42 #42
epochs= 100 #400
lr=0.01
weight_decay = 5e-4
hidden = 8
dropout= 0.5
num_heads=8
LRalpha=0.2
np.random.seed(seed)
torch.manual_seed(seed)
if use_cuda:
torch.cuda.manual_seed(seed)
flds=3
te_losses=[]
te_acc=[]
te_auc=[]
outputs=[]
te_f1=[]
cnn_acc=[]
cnn_f1=[]
fc_acc=[]
fc_f1=[]
fc_auc=[]
cnn_auc=[]
for f in range(len(pt_features)):
te_features = pt_te_features[f]
te_labels = pt_te_labels[f]
tr_f3atures = pt_features[f]
tr_lab3ls = pt_labels[f]
fact=[]
rk2=[]
sl=[]
tl=[]
for acq in tr_f3atures:
fact.append(acq[0])
rk2.append(acq[1])
sl.append(acq[2])
tl.append(acq[3])
fact=torch.stack(fact)
rk2=torch.stack(rk2)
sl=torch.stack(sl)
tl=torch.stack(tl)
te_fact=[]
te_rk2=[]
te_sl=[]
te_tl=[]
for acq in te_features:
te_fact.append(acq[0])
te_rk2.append(acq[1])
te_sl.append(acq[2])
te_tl.append(acq[3])
te_fact=torch.stack(te_fact)
te_rk2=torch.stack(te_rk2)
te_sl=torch.stack(te_sl)
te_tl=torch.stack(te_tl)
tr_algo=[fact,rk2,sl,tl]
te_algo=[te_fact,te_rk2,te_sl,te_tl]
algo_acc=[]
algo_f1=[]
algo_auc=[]
for algo in range(len(tr_algo)):
fcf1a,fcacca,fcauca =
fcNN(tr_algo[algo],tr_lab3ls,te_algo[algo],te_labels)
algo_acc.append(fcacca)
algo_f1.append(fcf1a)
algo_auc.append(fcauca)
fcacc = np.mean(algo_acc)
fcf1 = np.mean(algo_f1)
fcauc = np.mean(algo_auc)
fc_acc.append(fcacc)
fc_f1.append(fcf1)
fc_auc.append(fcauc)
print(f'Mean fcNN performce: Accuracy = {fcacc}, F1 = {fcf1:.0f}')
del tr_algo,te_algo,fact,rk2,sl,tl,te_fact,te_rk2,te_sl,te_tl
cnn = SimpleCNN()
f_acc,f_f1,cnnauc =
CNN(cnn, 32, 100, 0.01,tr_f3atures,tr_lab3ls,te_features,te_labels)
print(f'CNN Test Acc: {f_acc}')
print(f'CNN Test F1: {f_f1}')
cnn_acc.append(f_acc)
cnn_f1.append(f_f1)
cnn_auc.append(cnnauc)
model = GCNetwork(nfeat=tr_f3atures[0][0].shape[1],
nhid=hidden,
nclass=tr_lab3ls.max().item() + 1,
dropout=dropout,
nheads=num_heads,
alpha=LRalpha)
optimizer = optim.Adam(model.parameters(),
lr=lr, weight_decay=weight_decay)
v_losses = [0,0,0,0,0]
v_accs = []
tr_losses = []
tr_accs = []
tr_losses5=[0,0,0,0,0]
tr_features = tr_f3atures #[:int(0.9*len(tr_f3atures))]
tr_labels = tr_lab3ls #[:int(0.9*len(tr_f3atures))]
#val_features = tr_f3atures[int(0.9*len(tr_f3atures)):]
#val_labels = tr_lab3ls[int(0.9*len(tr_f3atures)):]
#Train model
t_total = time.time()
for epoch in range(epochs):
with open('attentions.txt','a') as attn:
attn.write(str(epoch))
tr_idx = np.random.permutation(len(tr_features))
tr_fold_lbls_sh=[]
tr_fold_ftrs_sh=[]
for tri in tr_idx:
tr_fold_ftrs_sh.append(tr_features[tri])
tr_fold_lbls_sh.append(tr_labels[tri])
tr_features = tr_fold_ftrs_sh
tr_labels = torch.stack(tr_fold_lbls_sh)
tr_loss=[]
tracc=[]
for batch in range(0,len(tr_features),32):
batch_features = tr_features[batch:batch+32]
batch_labels = tr_labels[batch:batch+32]
tloss,tacc =
train(epoch, adj, tr_features,
tr_labels, model, optimizer,f)
tr_loss.append(tloss.detach())
tracc.append(tacc.detach())
tr_losses.append(torch.mean(torch.stack(tr_loss)))
tr_accs.append(torch.mean(torch.stack(tracc))*100)
""" if using validation set, uncomment the lines below """
#if epoch%5==0:
# tr_losses5[4] = tr_losses5[3]
# tr_losses5[3] = tr_losses5[2]
# tr_losses5[2] = tr_losses5[1]
# tr_losses5[1] = tr_losses5[0]
# tr_losses5[0] = tloss#
# f_output,f_l,f_a=test(adj,val_features,val_labels,model)
# print(tloss.detach(),f_l)
# #v_losses[4] = v_losses[3]
# v_losses[3] = v_losses[2]
# v_losses[2] = v_losses[1]
# v_losses[1] = v_losses[0]
# v_losses[0] = f_l
# if v_losses[3] <= v_losses[2] and v_losses[2] <= v_losses[1] and v_losses[1] <= v_losses[0] and epoch>20:
# print(f'Ending optimization at epoch {epoch}')
# break
print("Optimization Finished!")
print("Total time elapsed: {:.4f}s".format(time.time() - t_total))
tst_fld_acc=[]
tst_fld_f1=[]
tst_fld_losses=[]
f_output,f_l,f_a,f_auc=test(adj,te_features,te_labels,model,f)
f_output = f_output.detach()
preds = f_output.max(1)[1].type_as(te_labels)
f1 = sklearn.metrics.f1_score(te_labels,preds)
te_auc.append(f_auc)
te_f1.append(f1)
te_acc.append(f_a.detach())
te_losses.append(f_l.detach())
#print('Test Losses: ',torch.stack(fld_te_losses).mean(),torch.stack(fld_te_losses).std())
#print('Test Accuracy :',torch.stack(fld_te_acc).mean(),torch.stack(fld_te_acc).std())
bl_te_acc['GCN']=te_acc
bl_te_f1['GCN']=te_f1
bl_te_auc['GCN']=te_auc
print('Test Losses: ',torch.stack(te_losses).mean(),torch.stack(te_losses).std())
print('Test Accuracy: ',torch.stack(te_acc).mean(),torch.stack(te_acc).std())
print('TesT F1: ',np.mean(te_f1),np.std(te_f1))
bl_te_acc['CNN']=cnn_acc
bl_te_f1['CNN']=cnn_f1
bl_te_acc['fcNN']=fc_acc
bl_te_f1['fcNN']=fc_f1
bl_te_auc['CNN']=cnn_auc
bl_te_auc['fcNN']=fc_auc
for f in range(len(skl_features)):
tr_features = skl_features[f]
tr_labels = skl_labels[f]
te_features = skl_te_features[f]
te_labels = [ls for ls in skl_te_labels[f]]
tst_acc, tst_f1, clf, tst_auc = baselineML(tr_features,tr_labels,
te_features,te_labels)
for i,c in enumerate(clf):
if c not in bl_te_acc.keys(): bl_te_acc[c]=[]
if c not in bl_te_f1.keys(): bl_te_f1[c]=[]
if c not in bl_te_auc.keys(): bl_te_auc[c]=[]
bl_te_acc[c].append(tst_acc[c])
bl_te_f1[c].append(tst_f1[c])
bl_te_auc[c].append(tst_auc[c])
with open('baselines.txt','a') as bl:
bl.write('------NEW RESULTS----')
for c in bl_te_acc.keys():
print(c)
print(f'Acc: {np.mean(bl_te_acc[c])}, +/-{np.std(bl_te_acc[c])}')
print(f'F1 : {np.mean(bl_te_f1[c])}, +/-{np.std(bl_te_f1[c])}')
print(f'AUC : {np.mean(bl_te_auc[c])}, +/-{np.std(bl_te_f1[c])}')
with open('baselines.txt','a') as bl:
bl.write(c)
bl.write(f'\tAcc: {bl_te_acc[c]}\t')
bl.write(f'F1: {bl_te_f1[c]}\t')
bl.write(f'AUC: {bl_te_auc[c]}')
bl.write('\n')