-
Notifications
You must be signed in to change notification settings - Fork 1
/
GCN_predict_train.py
277 lines (268 loc) · 15.7 KB
/
GCN_predict_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
from __future__ import division
from __future__ import print_function
import argparse
import matplotlib.pyplot as plt
import pandas as pd
import pickle
from models import *
from utils import *
import scipy.sparse as sp
from sklearn.metrics import roc_auc_score
import torch_geometric.transforms as T
from torch_geometric.datasets import Planetoid
from torch_geometric.utils import negative_sampling
import warnings
warnings.filterwarnings("ignore")
# Training settings
parser = argparse.ArgumentParser()
parser.add_argument('--no-cuda', action='store_true', default=False,
help='Disables CUDA training.')
parser.add_argument('--seed', type=int, default=42, help='Random seed.')
parser.add_argument('--epochs', type=int, default=400,
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-3,
help='Weight decay (L2 loss on parameters).')
parser.add_argument('--hidden', type=int, default=256,
help='Number of hidden units.')
parser.add_argument('--dropout', type=float, default=0.8,
help='Dropout rate (1 - keep probability).')
parser.add_argument("--normalization", default="FirstOrderGCN", # FirstOrderGCN
help="The normalization on the adj matrix.")
parser.add_argument('--dataset', default="cora", help="The data set") # citeseer cora
parser.add_argument('--datapath', default="./data", help="The data path.")
parser.add_argument('--lradjust', action='store_true',
default=True, help='Enable leraning rate adjust.(ReduceLROnPlateau or Linear Reduce)')
OUTPUT_PATH = r'./predict/'
test_flag = True
args = parser.parse_args()
args.cuda = torch.cuda.is_available()
# random seed setting
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
# def sparse_mx_to_torch_sparse_tensor(sparse_mx):
# sparse_mx = sparse_mx.tocoo().astype(np.float32)
# indices = torch.from_numpy(
# np.vstack((sparse_mx.row, sparse_mx.col)).astype(np.int64))
# values = torch.from_numpy(sparse_mx.data)
# shape = torch.Size(sparse_mx.shape)
# return torch.sparse.FloatTensor(indices, values, shape)
@torch.no_grad()
def eval_link_predictor(model, data, adj_test):
model.eval()
z = model.encode(data.x, adj_test)
out = model.decode(z, data.edge_label_index).view(-1).sigmoid()
return roc_auc_score(data.edge_label.cpu().numpy(), out.cpu().numpy())
def get_lr(optimizer):
for param_group in optimizer.param_groups:
return param_group['lr']
# define the training function.
def train(model, train_data, adj_train_nrom, val_data):
model.train()
optimizer.zero_grad()
z = model.encode(train_data.x, adj_train_nrom)
neg_edge_index = negative_sampling(
edge_index=train_data.edge_index, num_nodes=train_data.num_nodes,
num_neg_samples=train_data.edge_label_index.size(1), method='sparse')
edge_label_index = torch.cat(
[train_data.edge_label_index, neg_edge_index],
dim=-1,
)
edge_label = torch.cat([
train_data.edge_label,
train_data.edge_label.new_zeros(neg_edge_index.size(1))
], dim=0)
out = model.decode(z, edge_label_index).view(-1)
loss = criterion(out, edge_label)
loss.backward()
optimizer.step()
val_auc = eval_link_predictor(model, val_data, adj_train_nrom)
return loss, val_auc, get_lr(optimizer)
def test(model, test_data, adj_test_nrom):
model.eval()
test_auc = eval_link_predictor(model, test_data, adj_test_nrom)
return test_auc
if not test_flag:
test_cases = [
# base
{'num_layers':2, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
# test layer 1
{'num_layers':3, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':4, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':8, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':16, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':32, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
# test self loops 6
{'num_layers':2, 'add_self_loops':True, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':4, 'add_self_loops':True, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':8, 'add_self_loops':True, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':16, 'add_self_loops':True, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':32, 'add_self_loops':True, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
# test batch norm 11
{'num_layers':2, 'add_self_loops':False, 'add_bn':True, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':4, 'add_self_loops':False, 'add_bn':True, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':8, 'add_self_loops':False, 'add_bn':True, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':16, 'add_self_loops':False, 'add_bn':True, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':32, 'add_self_loops':False, 'add_bn':True, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'linear'},
# test use_pairnorm 16
{'num_layers':2, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'PN', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':4, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'PN', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':8, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'PN', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':16, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'PN', 'drop_edge':1.0, 'activation':'linear'},
{'num_layers':32, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'PN', 'drop_edge':1.0, 'activation':'linear'},
# test drop_edge 21
{'num_layers':2, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.8, 'activation':'linear'},
{'num_layers':4, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.8, 'activation':'linear'},
{'num_layers':8, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.8, 'activation':'linear'},
{'num_layers':16, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.8, 'activation':'linear'},
{'num_layers':32, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.8, 'activation':'linear'},
{'num_layers':2, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.6, 'activation':'linear'},
{'num_layers':4, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.6, 'activation':'linear'},
{'num_layers':8, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.6, 'activation':'linear'},
{'num_layers':16, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.6, 'activation':'linear'},
{'num_layers':32, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.6, 'activation':'linear'},
{'num_layers':2, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.4, 'activation':'linear'},
{'num_layers':4, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.4, 'activation':'linear'},
{'num_layers':8, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.4, 'activation':'linear'},
{'num_layers':16, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.4, 'activation':'linear'},
{'num_layers':32, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':0.4, 'activation':'linear'},
# test activation 36
{'num_layers':2, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'tanh'},
{'num_layers':4, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'tanh'},
{'num_layers':8, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'tanh'},
{'num_layers':16, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'tanh'},
{'num_layers':32, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'tanh'},
{'num_layers':2, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'relu'},
{'num_layers':4, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'relu'},
{'num_layers':8, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'relu'},
{'num_layers':16, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'relu'},
{'num_layers':32, 'add_self_loops':False, 'add_bn':False, 'use_pairnorm':'None', 'drop_edge':1.0, 'activation':'relu'},
]
else:
test_cases = [
{'num_layers':8, 'add_self_loops':True, 'add_bn':True, 'use_pairnorm':'PN', 'drop_edge':1.0, 'activation':'linear'},
]
dataset = Planetoid(root=args.datapath, name=args.dataset)
graph = dataset[0]
del graph.train_mask
del graph.val_mask
del graph.test_mask
split = T.RandomLinkSplit(
num_val=0.1,
num_test=0.1,
is_undirected=True,
add_negative_train_samples=False,
neg_sampling_ratio=1.0,
)
train_data, val_data, test_data = split(graph)
adj_train = sp.csr_matrix(
(np.ones(train_data.num_edges), (train_data.edge_index[0, :], train_data.edge_index[1, :])),
shape=[train_data.num_nodes, train_data.num_nodes])
adj_train_nrom = adj_normalize(adj_train)
adj_train_nrom = sparse_mx_to_torch_sparse_tensor(adj_train_nrom)
adj_test = sp.csr_matrix((np.ones(test_data.num_edges), (test_data.edge_index[0, :], test_data.edge_index[1, :])),
shape=[test_data.num_nodes, test_data.num_nodes])
adj_test_nrom = adj_normalize(adj_test)
adj_test_nrom = sparse_mx_to_torch_sparse_tensor(adj_test_nrom)
# convert to cuda
train_data = train_data.cuda()
val_data = val_data.cuda()
test_data = test_data.cuda()
adj_train_nrom = adj_train_nrom.cuda()
adj_test_nrom = adj_test_nrom.cuda()
for i_case, kwargs in enumerate(test_cases):
args.nbaseblocklayer = kwargs['num_layers']
args.withloop = kwargs['add_self_loops']
args.withbn = kwargs['add_bn']
args.use_pairnorm = kwargs['use_pairnorm']
args.sampling_percent = kwargs['drop_edge']
args.activation = kwargs['activation']
nfeat = dataset.num_node_features
nclass = dataset.num_classes
model = LinkNet(nfeat=nfeat,
nhid=args.hidden,
nclass=nclass,
dropout=args.dropout,
self_loops=args.withloop,
num_layers=args.nbaseblocklayer,
norm_mode=args.withbn,
use_pairnorm=args.use_pairnorm, # 'None', 'PN', 'PN-SI', 'PN-SCS'
activation=args.activation) # relu linear
model = model.cuda()
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr) # lr=1e-3 注意：在2.17日15.20之前的数据的学习率均固定为1e-3
# optimizer = torch.optim.Adam(model.parameters(), lr=args.lr, weight_decay=args.weight_decay)
if args.lradjust:
# scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, patience=50, factor=0.618)
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[150, 250, 350], gamma=0.5)
criterion = torch.nn.BCEWithLogitsLoss()
# Train model
loss_train = np.zeros((args.epochs,))
auc_val = np.zeros((args.epochs,))
lr = np.zeros((args.epochs,))
best_auc =0
for epoch in range(args.epochs):
loss, val_auc, lr_get = train(model, train_data, adj_train_nrom, val_data)
print("loss:" + str(loss.item()) + " "+"val_auc:" + str(val_auc.item())+" "+"lr:"+str(lr_get))
loss_train[epoch], auc_val[epoch], lr[epoch] = loss, val_auc, lr_get
if test_flag:
if best_auc < val_auc:
best_auc = val_auc
torch.save(model.state_dict(), OUTPUT_PATH + 'checkpoint-best-auc.pkl')
# Testing
if test_flag:
model.load_state_dict(torch.load(OUTPUT_PATH + 'checkpoint-best-auc.pkl'))
test_auc = test(model, test_data, adj_test_nrom)
print("%i\t%.6f\t%.6f\t%.6f\t%.6f" % (
i_case, lr[-1], loss_train[-1], auc_val[-1], test_auc))
# else:
kwargs['best_auc'] = max(auc_val)
args.nbaseblocklayer = kwargs['num_layers']
args.withloop = kwargs['add_self_loops']
args.withbn = kwargs['add_bn']
args.sampling_percent = kwargs['drop_edge']
args.use_pairnorm = kwargs['use_pairnorm']
file_path=OUTPUT_PATH
# nameloss = 'case%sLr%floss_layer%iselfloop%iwithbn%iuse_pairnorm%sdrop_edge%factivation%s' % (
# i_case, lr[-1], args.nbaseblocklayer, args.withloop, args.withbn, args.use_pairnorm, args.sampling_percent,
# args.activation)
# nameacc = 'case%sLr%facc_layer%iselfloop%iwithbn%iuse_pairnorm%sdrop_edge%factivation%s' % (
# i_case, lr[-1], args.nbaseblocklayer, args.withloop, args.withbn, args.use_pairnorm, args.sampling_percent,
# args.activation)
plt.plot(list(range(args.epochs)), loss_train, label='loss_train')
plt.xlabel('epoch')
plt.ylabel('loss')
plt.title('train loss VS val loss')
plt.legend()
plt.savefig(file_path + "loss.jpg")
# plt.show()
plt.clf()
plt.plot(list(range(args.epochs)), auc_val, label='auc_val')
plt.xlabel('epoch')
plt.ylabel('auc')
plt.title('train auc VS val auc')
plt.legend()
plt.savefig(file_path + "auc.jpg")
plt.show()
# pkl_name1 = 'task%icase%iLr%flayer%iselfloop%iwithbn%iuse_pairnorm%sdrop_edge%factivation%sloss_train.pkl' % (
# task, i_case, lr[-1], args.nbaseblocklayer, args.withloop, args.withbn, args.use_pairnorm, args.sampling_percent,
# args.activation)
# with open(pkl_name1, 'wb') as f:
# pickle.dump(loss_train, f)
# pkl_name2 = 'task%icase%iLr%flayer%iselfloop%iwithbn%iuse_pairnorm%sdrop_edge%factivation%sacc_val.pkl' % (
# task, i_case, lr[-1], args.nbaseblocklayer, args.withloop, args.withbn, args.use_pairnorm, args.sampling_percent,
# args.activation)
# with open(pkl_name2, 'wb') as f:
# pickle.dump(auc_val, f)
# pkl_name3 = 'task%icase%iLr%flayer%iselfloop%iwithbn%iuse_pairnorm%sdrop_edge%factivation%sacc_val.pkl' % (
# task, i_case, lr[-1], args.nbaseblocklayer, args.withloop, args.withbn, args.use_pairnorm, args.sampling_percent,
# args.activation)
# with open(pkl_name3, 'wb') as f:
# pickle.dump(lr, f)
# if not test_flag:
# if task == 1:
# pd.DataFrame(test_cases).to_csv(f'{args.dataset1}-Result1.csv')
# else:
# pd.DataFrame(test_cases).to_csv(f'{args.dataset2}-Result11.csv')