forked from hedongxiao-tju/SGRL
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
185 lines (146 loc) · 7.17 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
import numpy as np
import torch_geometric.transforms as T
from torch_geometric.datasets import Planetoid, CitationFull, WikiCS, Amazon, Coauthor, WebKB, Actor, WikipediaNetwork
from sklearn import metrics
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import GridSearchCV, ShuffleSplit, train_test_split
from sklearn.multiclass import OneVsRestClassifier
from sklearn.preprocessing import OneHotEncoder, normalize
from data import load_dataset
from models import Conv, Online, Target
import torch
import random
import torch_geometric
import copy
import time
from tqdm import tqdm
from torch_geometric.utils import train_test_split_edges
import warnings
import argparse
import sys
import os
def adj_norm(adj_t):
deg = torch.sparse.sum(adj_t, dim=1).to_dense()
deg_inv_sqrt = deg.pow(-0.5)
adj_t = adj_t * deg_inv_sqrt.view(1, -1)
adj_t = adj_t * deg_inv_sqrt.view(-1, 1)
return adj_t
def fit_logistic_regression(X, y, data_random_seed=1, repeat=3):
one_hot_encoder = OneHotEncoder(categories='auto')
y = one_hot_encoder.fit_transform(y.reshape(-1, 1)).astype(bool)
X = normalize(X, norm='l2')
rng = np.random.RandomState(data_random_seed)
accuracies = []
for _ in range(repeat):
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.8, random_state=rng)
logreg = LogisticRegression(solver='liblinear')
c = 2.0 ** np.arange(-10, 11)
cv = ShuffleSplit(n_splits=5, test_size=0.5)
clf = GridSearchCV(estimator=OneVsRestClassifier(logreg), param_grid=dict(estimator__C=c),
n_jobs=5, cv=cv, verbose=0)
clf.fit(X_train, y_train)
y_pred = clf.predict_proba(X_test)
y_pred = np.argmax(y_pred, axis=1)
y_pred = one_hot_encoder.transform(y_pred.reshape(-1, 1)).astype(bool)
test_acc = metrics.accuracy_score(y_test, y_pred)
accuracies.append(test_acc)
return accuracies
def train_online(online,optimizer,data):
online.train()
h,h_pred,h_target = online(data.x,data.edge_index)
loss = online.get_loss(h_pred,h_target.detach())
loss.backward()
optimizer.step()
online.update_target_encoder()
return loss.item()
def train_target(target,optimizier,data):
target.train()
h_target = target(data.x,data.edge_index)
loss = target.get_loss(h_target)
loss.backward()
optimizier.step()
return loss.item()
def run(args):
log_dir = args.log_dir
path = args.data_dir
with open(log_dir, 'a') as f:
f.write(str(args))
f.write('\n\n\n')
trials = args.trials
torch_geometric.seed.seed_everything(args.seed)
seed = args.seed
for trial in range(trials):
e1_lr = args.e1_lr
e2_lr = args.e2_lr
weight_decay = args.weight_decay
hidden_dim = args.hidden_dim
activation = torch.nn.PReLU()
num_layers = args.num_layers
num_epochs = args.num_epochs
momentum = args.momentum
dataset_name = args.dataset_name
dataset = load_dataset(dataset_name,path)
data = dataset[0]
device = torch.device('cuda')
data = data.to(device)
num_hop = args.num_hop
nb_nodes = data.x.size()[0]
self_loop_for_adj = torch.Tensor([i for i in range(nb_nodes)]).unsqueeze(0)
self_loop_for_adj = torch.concat([self_loop_for_adj, self_loop_for_adj], dim=0)
slsp_adj = torch.concat([data.edge_index.to('cpu'), self_loop_for_adj], dim=1)
slsp_adj = torch.sparse.FloatTensor(slsp_adj.long(), torch.ones(slsp_adj.size()[1]),
torch.Size([nb_nodes, nb_nodes]))
slsp_adj = adj_norm(slsp_adj).to(device)
online_conv = Conv(data.x.size()[1],hidden_dim,hidden_dim,activation,num_layers).to(device)
target_conv = Conv(data.x.size()[1],hidden_dim,hidden_dim,activation,num_layers).to(device)
online_model = Online(online_conv,target_conv,hidden_dim,slsp_adj,num_hop,momentum).to(device)
target_model = Target(target_conv).to(device)
online_optimizer = torch.optim.Adam(online_model.parameters(),lr=e1_lr)
target_optimizer = torch.optim.Adam(target_model.parameters(),lr=e2_lr)
best_online_loss = 1e9
best_target_loss = 1e9
tag = dataset_name + '_' + str(time.time())
with tqdm(total = num_epochs,desc='(T)') as pbar:
for epoch in range(num_epochs):
online_optimizer.zero_grad()
target_optimizer.zero_grad()
online_loss = train_online(online_model,online_optimizer,data)
if online_loss < best_online_loss:
best_online_loss = online_loss
torch.save(online_model.state_dict(), 'pkl/pkl_online/best_online' + tag + '.pkl')
target_loss = train_target(target_model,target_optimizer,data)
if target_loss < best_target_loss:
best_target_loss_loss = target_loss
torch.save(target_model.state_dict(), 'pkl/pkl_target/best_target' + tag + '.pkl')
target_model.load_state_dict(torch.load('pkl/pkl_target/best_target' + tag + '.pkl'))
pbar.set_postfix({'loss': online_loss})
pbar.update()
online_model.load_state_dict(torch.load('pkl/pkl_online/best_online' + tag + '.pkl'))
online_model.eval()
or_embeds, pr_embeds = online_model.embed(data.x,data.edge_index,slsp_adj,num_hop)
embeds = or_embeds + pr_embeds
scores = fit_logistic_regression(embeds.detach().cpu().numpy(), data.y.cpu().numpy())
m = np.mean(scores)
n = np.var(scores)
print(m,n)
with open(log_dir, 'a') as f:
f.write('sgrl_mean: '+str(m)[0:7]+' sgrl_std: '+ str(n))
f.write('\n')
if __name__ == '__main__':
warnings.filterwarnings("ignore")
parser = argparse.ArgumentParser('SGRL')
parser.add_argument('--dataset_name', type=str, default='Photo', help='dataset_name')
parser.add_argument('--data_dir', type=str, default='../../datasets', help='data_dir')
parser.add_argument('--log_dir', type=str, default='./log/log_Photo', help='log_dir')
parser.add_argument('--e1_lr', type=float, default=0.001, help='online_learning_rate')
parser.add_argument('--e2_lr', type=float, default=0.001, help='target_learning_rate')
parser.add_argument('--momentum', type=float, default=0.99, help='EMA')
parser.add_argument('--weight_decay', type=float, default=0., help='weight_decay')
parser.add_argument('--num_epochs', type=int, default=700, help='num_epochs')
parser.add_argument('--seed', type=int, default=66666, help='seed')
parser.add_argument('--hidden_dim', type=int, default=1024, help='hidden_dim')
parser.add_argument('--num_layers', type=int, default=1, help='num_layers')
parser.add_argument('--num_hop', type=int, default=1, help='num_hop')
parser.add_argument('--trials', type=int, default=20, help='trials')
args = parser.parse_args()
run(args)