forked from pkuyzy/TLC-GNN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpipelines_GIN.py
150 lines (145 loc) · 6.18 KB
/
pipelines_GIN.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
import os
os.environ['CUDA_VISIBLE_DEVICES']="9"
import torch
import loaddatas as lds
import torch.nn.functional as F
import random
import numpy as np
#from baselines import ConvCurv,ConvCurv_pers
#from baselines import ConvCurv_perslay as ConvCurv_pers
#from learnable_filter import ConvCurv as ConvCurv_pers
#from learnable_filter import ConvCurv_GIN as ConvCurv_pers
from Knowledge_Distillation import ConvCurv_GIN as ConvCurv_pers
#load the neural networks
def train(train_mask):
model.train()
#for parameters in model.modelGIN.lin1.parameters():
# print(parameters)
optimizer.zero_grad()
F.nll_loss(model(data)[train_mask], data.y[train_mask]).backward()
optimizer.step()
def test(train_mask,val_mask,test_mask):
model.eval()
logits, accs = model(data), []
for mask in [train_mask, val_mask, test_mask]:
pred = logits[mask].max(1)[1]
acc = pred.eq(data.y[mask]).sum().item() / mask.sum().item()
accs.append(acc)
accs.append(F.nll_loss(model(data)[val_mask], data.y[val_mask]))
return accs
#load dataset
times=range(10)
wait_total=100
total_epochs = 200
pipelines=['ConvCurv_pers']
pipeline_acc={'ConvCurv_pers':[i for i in times]}
pipeline_acc_sum={'ConvCurv_pers':0}
d_names=[#'Cora',
#'Citeseer',
'PubMed',
#'Photo'
#'Physics',
#'CS',
#'Computers'
]
#d_names = ['Photo','CS','Physics','Computers']
#d_names=['Computers']
for d_name in d_names:
if d_name == 'Photo' or d_name == 'Computers':
wait_total = 200
total_epochs = 500
else:
wait_total = 100
total_epochs = 200
f2=open('scores/pipe_benchmark_' +d_name+ '_GIN_new.txt', 'w+')
f2.write('{0:7} {1:7}\n'.format(d_name,'ConvCurv'))
f2.flush()
if d_name=='Cora' or d_name=='Citeseer' or d_name=='PubMed':
d_loader='Planetoid'
elif d_name=='Computers' or d_name=='Photo':
d_loader='Amazon'
elif d_name == 'CS' or d_name == 'Physics':
d_loader='Coauthor'
else:
d_loader = 'Synthesis'
dataset=lds.loaddatas(d_loader,d_name)
#already generated, no need to generate
'''
if not os.path.exists('./data/curvature/graph_'+d_name+'.edge_list'):
print("start writing ricci curvature")
Gd = nx.Graph()
ricci_edge_index_ = np.array(dataset[0].edge_index)
ricci_edge_index = [(ricci_edge_index_[0, i], ricci_edge_index_[1, i]) for i in
range(np.shape(dataset[0].edge_index)[1])]
Gd.add_edges_from(ricci_edge_index)
Gd_OT = OllivierRicci(Gd, alpha=0.5, method="OTD", verbose="INFO")
print("adding edges finished")
Gd_OT.compute_ricci_curvature()
ricci_list = []
for n1, n2 in Gd_OT.G.edges():
ricci_list.append([n1, n2, Gd_OT.G[n1][n2]['ricciCurvature']])
ricci_list.append([n2, n1, Gd_OT.G[n1][n2]['ricciCurvature']])
ricci_list = sorted(ricci_list)
ricci_file = open('./data/curvature/graph_'+d_name+'.edge_list','w')
for ricci_i in range(len(ricci_list)):
ricci_file.write(str(ricci_list[ricci_i][0]) + " " + str(ricci_list[ricci_i][1]) + " " + str(ricci_list[ricci_i][2]) + "\n")
ricci_file.close()
print("writing ricci curvature finished")
'''
for time in times:
for Conv_method in pipelines:
if d_loader != 'Synthesis':
data=dataset[0]
else:
data = dataset[0]
data.x = data.x[:, :10]
data.x = torch.ones(data.x.size())
index=[i for i in range(len(data.y))]
if d_loader == 'Coauthor' or d_loader == 'Amazon':
train_len=20*int(data.y.max()+1)
train_mask=torch.tensor([i < train_len for i in index])
val_mask=torch.tensor([i >= train_len and i < 500+train_len for i in index])
test_mask=torch.tensor([i >= len(data.y)-1000 for i in index])
elif d_loader == 'Planetoid':
train_mask=data.train_mask.bool()
val_mask=data.val_mask.bool()
test_mask=data.test_mask.bool()
else:
random.shuffle(index)
len_mul = int(1000 / 5)
train_mask = torch.tensor([i < len_mul * 2 for i in index])
val_mask = torch.tensor([(i >= len_mul * 2) and (i < len_mul * 4) for i in index])
test_mask = torch.tensor([i >= (len(data.y) - len_mul) for i in index])
model,data = locals()[Conv_method].call(data,dataset.name,data.x.size(1),dataset.num_classes)
optimizer = torch.optim.Adam(model.parameters(), lr=0.005, weight_decay=0.0005)
best_val_acc = test_acc = 0.0
best_val_loss = np.inf
wait_step = 0
for epoch in range(1, total_epochs+1):
train(train_mask)
train_acc,val_acc,tmp_test_acc,val_loss = test(train_mask,val_mask,test_mask)
if val_acc>=best_val_acc:
test_acc=tmp_test_acc
best_val_acc=val_acc
best_val_loss=val_loss
wait_step=0
else:
wait_step += 1
if wait_step == wait_total:
print('Early stop! Min loss: ', best_val_loss, ', Max accuracy: ', best_val_acc)
break
del model
del data
pipeline_acc[Conv_method][time]=test_acc
pipeline_acc_sum[Conv_method]=pipeline_acc_sum[Conv_method]+test_acc/len(times)
log ='Epoch: ' + str(total_epochs) + ', dataset name: '+ d_name + ', Method: '+ Conv_method + ' Test: {:.4f} \n'
print((log.format(pipeline_acc[Conv_method][time])))
f2.write('{0:4d} {1:4f}\n'.format(time,pipeline_acc[Conv_method][time]))
f2.flush()
f2.write('{0:4} {1:4f}\n'.format('std',np.std(pipeline_acc[Conv_method])))
f2.write('{0:4} {1:4f}\n'.format('mean',np.mean(pipeline_acc[Conv_method])))
f2.close()
# delete evaluation of other models
save_GIN_PI = "/data1/curvGN_LP/data/data/KD/predicted/" + d_name + "_temp.pt"
if os.path.exists(save_GIN_PI):
os.remove(save_GIN_PI)