This repository has been archived by the owner on Jun 18, 2023. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgrape_model.py
164 lines (135 loc) · 6.02 KB
/
grape_model.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
from __future__ import division
from __future__ import print_function
import time
from motif_search import *
from utils import *
from models import GRAPE
import setproctitle
import os
import scipy.sparse as sp
import random
import torch
import torch.nn.functional as F
import torch.optim as optim
import search
import argparse
def get_parser():
parser = argparse.ArgumentParser()
# 'cite|cora', 'cite|citeseer', 'amazon', 'social|Amherst', 'social|Hamilton', 'social|Rochester', 'social|Lehigh', 'social|Johns Hopkins'
parser.add_argument('--data', default='cite|cora')
parser.add_argument('--gpu', default='0')
parser.add_argument('--lr', default=0.003)
parser.add_argument('--wd', default=0.00003)
parser.add_argument('--dropout', default=0.5)
parser.add_argument('--hid', default=32)
return parser
parser = get_parser()
args = parser.parse_args()
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
attn = True # Switch for squeeze-and-excite net
flag_acc = True # Accumulate motif count or not
model_name = 'GRAPE'
setproctitle.setproctitle(model_name)
# set random seed
seed = 42
np.random.seed(seed)
torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed(seed)
# grape model hyperparameter setting
num_genes = 5
early_stopping = 50
nepoch = 500
nlayer = 2
test_run = 10
# compute accuracy and loss of the trained models
def evaluate(pred, target, idx):
pred = F.log_softmax(pred, dim=1)
loss = F.nll_loss(pred[idx], target[idx])
acc = accuracy(pred[idx], target[idx]).item()
return loss, acc
def train_model(nlayer, nepoch, candidate_adj, features, labels, idx_train, idx_val, idx_test, attn, lr, weight_decay, dropout, hidden):
# flatten the ADJ of different motifs and add in a self-loop
ngene = len(candidate_adj)
nrole = [len(item) for item in candidate_adj]
nclass = labels.max().item() + 1
model = GRAPE(nfeat=features.shape[1], nlayer=nlayer, nhid=hidden, nclass=nclass, nrole=nrole, ngene=ngene, dropout=dropout, attn=attn)
cur_lr = lr
optimizer = optim.Adam(model.parameters(), lr=cur_lr, weight_decay=weight_decay)
if torch.cuda.is_available():
model.cuda()
features = features.cuda()
candidate_adj = [[itemtemp.cuda() for itemtemp in temp] for temp in candidate_adj]
labels = labels.cuda()
idx_train = idx_train.cuda()
idx_val = idx_val.cuda()
idx_test = idx_test.cuda()
loss_val_list = []
# Train model
t_total = time.time()
for epoch in range(nepoch):
# Construct feed dictionary
model.train()
optimizer.zero_grad()
output = model(features, candidate_adj)
loss_train, acc_train = evaluate(output, labels, idx_train)
loss_train.backward()
optimizer.step()
# Evaluate validation set performance separately,
# deactivates dropout during validation run.
model.eval()
output = model(features, candidate_adj)
loss_val, acc_val = evaluate(output, labels, idx_val)
loss_val_list.append(loss_val.item())
if epoch%10==1:
print('Epoch: {:04d}'.format(epoch+1), 'loss_train: {:.4f}'.format(loss_train.item()), 'acc_train: {:.4f}'.format(acc_train),
'loss_val: {:.4f}'.format(loss_val.item()),'acc_val: {:.4f}'.format(acc_val))
if epoch%100==99:
cur_lr = 0.5 * cur_lr
for param_group in optimizer.param_groups:
param_group['lr'] = cur_lr
if epoch > 200 and loss_val_list[-1] > np.mean(loss_val_list[-(early_stopping+1):-1]):
break
# Test model
model.eval()
output = model(features, candidate_adj)
loss_test, acc_test = evaluate(output, labels, idx_test)
print("Train accuracy= {:.4f}".format(acc_train), "Val accuracy= {:.4f}".format(acc_val), "Test accuracy= {:.4f}".format(acc_test), "epoch= {:04d}".format(epoch))
return acc_test
adj, features, labels, idx_train, idx_val, idx_test, flag_direct, population_test, select_index = read_data(args.data)
# Initialize incsearch and the motif adj matrix
search_base = np.array(adj.toarray(),dtype=np.int32) # dense array of base adj
print('Dataset contains:',len(search_base),'nodes,', sum(sum(search_base)), 'edges.')
node_num = len(search_base)
search.init_incsearch_model(search_base, flag_direct, flag_acc)
adj_dic = {}
init_motif = np.zeros((2, 2), dtype=np.int32)
# adj = normalize(adj)
if flag_direct:
init_motif[1, 0] = 1
adj_dic[str(list(init_motif.flatten()))] = [sparse_mx_to_torch_sparse_tensor(sp.eye(node_num)), sparse_mx_to_torch_sparse_tensor(adj)] # self-loop
init_motif[0, 1] = 1
init_motif[1, 0] = 0
adj_dic[str(list(init_motif.flatten()))] = [sparse_mx_to_torch_sparse_tensor(sp.eye(node_num)), sparse_mx_to_torch_sparse_tensor(adj.T)]
else:
init_motif[0, 1] = 1
init_motif[1, 0] = 1
adj_dic[str(list(init_motif.flatten()))] = [sparse_mx_to_torch_sparse_tensor(sp.eye(node_num)), sparse_mx_to_torch_sparse_tensor(adj)]
motifadj_test, adj_dic = construct_motif_adj_batch([population_test], adj_dic, search_base, flag_direct, flag_acc)
motifadj_test = motifadj_test[0]
motifadj_test = [motifadj_test[ind] for ind in select_index]
test_score = []
for ind in range(test_run):
id_list = range(node_num)
random.shuffle(id_list)
id_len = len(id_list)
idx_train = id_list[:int(id_len*0.6)]
idx_val = id_list[int(id_len*0.6):int(id_len*0.8)]
idx_test = id_list[int(id_len*0.8):]
idx_train = torch.LongTensor(idx_train)
idx_val = torch.LongTensor(idx_val)
idx_test = torch.LongTensor(idx_test)
acc = train_model(nlayer, nepoch, motifadj_test, features, labels, idx_train, idx_val, idx_test, attn, float(args.lr), float(args.wd), float(args.dropout), int(args.hid))
test_score.append(acc)
test_acc_mean, test_acc_std = np.mean(test_score), np.std(test_score)
print('Final result:', test_acc_mean, test_acc_std)