-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_cold.py
136 lines (102 loc) · 5.81 KB
/
train_cold.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
import os.path
import pickle
import torch
from dataset import read_pickle,mydatalist,prepare_twosides,twosides_pkl_loader,load_pkl
from torch_geometric.data import Data
from sklearn.metrics import roc_auc_score, precision_score, f1_score,recall_score
from sklearn.metrics import precision_recall_curve,auc,accuracy_score,average_precision_score
import numpy as np
from tqdm import tqdm
from util import log_util
from torch import optim
from DDI import DD_Pre
import time
import faulthandler
faulthandler.enable()
device = 'cuda' if torch.cuda.is_available() else 'cpu'
device_ids = range(torch.cuda.device_count())
torch.multiprocessing.set_sharing_strategy('file_system')
def test_DDI(test_loader, model):
model.eval()
y_pred = []
y_label = []
with torch.no_grad():
bar = tqdm(test_loader, ncols=80)
for i,batches in enumerate(bar):
head_list, tail_list, rel_list, Label = [data.to(device) for data in batches]
predictions= model(head_list, tail_list, rel_list,False)
predictions = predictions.squeeze()
predictions = torch.sigmoid(predictions)
predictions = predictions.detach().cpu().numpy()
Label = Label.detach().cpu().numpy()
y_label = y_label + Label.flatten().tolist()
y_pred = y_pred + predictions.flatten().tolist()
y_pred1 = np.array(y_pred)
y_label1 = np.array(y_label)
y_pred1_label = (y_pred1>=0.5).astype(np.int32)
roc_test_ACC,roc_test_AUROC,f1,roc_test_Pre,recall,roc_test_AUPR = accuracy_score(y_label1,y_pred1_label),roc_auc_score(y_label, y_pred1),f1_score(y_label1,y_pred1_label), precision_score(y_label1, y_pred1_label),recall_score(y_label1, y_pred1_label),average_precision_score(y_label1,y_pred1,average='micro')
p, r, t = precision_recall_curve(y_label1, y_pred1)
roc_test_AUC = auc(r, p)
return roc_test_ACC,roc_test_AUC,f1,roc_test_Pre,recall,roc_test_AUPR,roc_test_AUROC
if __name__ == '__main__':
dataset = 'twosides'
logs = log_util(dataset,'0.2')
for fold in [1]:
cold_train_dir = './cold_train.txt'
cold_c2_dir = './cold_test_C2.txt'
cold_c3_dir = './cold_test_C3.txt'
cold_train_data = prepare_twosides(cold_train_dir)
cold_c2_data = prepare_twosides(cold_c2_dir)
cold_c3_data = prepare_twosides(cold_c3_dir)
if torch.cuda.is_available():
model = DD_Pre(45,0.2,0.2).cuda()
train_loader = twosides_pkl_loader(cold_train_data,batch_size=2048,shuffle=True,num_workers=2,pin_memory = True)
c2 = twosides_pkl_loader(cold_c2_data,batch_size = 1024,shuffle=False,num_workers=2)
c3 = twosides_pkl_loader(cold_c3_data,batch_size = 1024,shuffle=False,num_workers=2)
optimizer = optim.Adam(model.parameters(), lr=1e-4, betas=(0.9, 0.99))
loss_history = []
t_total=time.time()
epochs=81
for epoch in range(0,epochs):
model.train()
t = time.time()
y_pred_train = []
y_label_train = []
bar = tqdm(train_loader,ncols=80)
total_loss = 0
batch = 0
for i, batches in enumerate(bar):
bar.set_description('Epoch ' + str(epoch))
head_list, tail_list, rel_list,Label = [data.to(device) for data in batches]
predictions, dis_loss = model(head_list, tail_list, rel_list,True)
predictions = predictions.squeeze()
loss1 = torch.nn.BCEWithLogitsLoss(reduction='sum')(predictions, Label)
loss = loss1
optimizer.zero_grad()
loss1.backward()
optimizer.step()
predictions = torch.sigmoid(predictions)
predictions = predictions.detach().cpu().numpy()
Label = Label.detach().cpu().numpy()
y_label_train = y_label_train + Label.flatten().tolist()
y_pred_train = y_pred_train + predictions.flatten().tolist()
total_loss += loss.item()
batch = len(y_label_train)
bar.set_postfix(loss ='%.5f' %(total_loss/batch))
y_pred_train = np.array(y_pred_train)
y_pred_train_label = (y_pred_train>=0.5).astype(np.int32)
y_label_train = np.array(y_label_train)
roc_train_ACC, roc_train_AUROC, train_f1, roc_train_Pre, train_recall, roc_train_AUPR = accuracy_score(y_label_train,y_pred_train_label),roc_auc_score(y_label_train, y_pred_train),f1_score(y_label_train,y_pred_train_label), precision_score(y_label_train, y_pred_train_label),recall_score(y_label_train, y_pred_train_label),average_precision_score(y_label_train,y_pred_train,average='micro')
p,r,t = precision_recall_curve(y_label_train,y_pred_train)
roc_train_AUC = auc(r,p)
print(roc_train_AUC)
logs.save_log(epoch,roc_train_ACC, roc_train_AUC, train_f1, roc_train_Pre, train_recall, roc_train_AUPR,roc_train_AUROC,'train',model,optimizer)
if epoch % 2 == 0:
roc_test_ACC, roc_test_AUC, f1, roc_test_Pre, recall, roc_test_AUPR,roc_test_AUROC = test_DDI(c2, model)
logs.save_log(epoch,roc_test_ACC,roc_test_AUC,f1,roc_test_Pre,recall,roc_test_AUPR,roc_test_AUROC,'c2',model,optimizer)
print(roc_test_Pre)
roc_test_ACC, roc_test_AUC, f1, roc_test_Pre, recall, roc_test_AUPR,roc_test_AUROC = test_DDI(c3, model)
logs.save_log(epoch,roc_test_ACC,roc_test_AUC,f1,roc_test_Pre,recall,roc_test_AUPR,roc_test_AUROC,'c3',model,optimizer)
print(roc_test_Pre)
torch.save(model.state_dict(),
'./save/twosides/{}_checkpoint.pt'.format('last'))