-
Notifications
You must be signed in to change notification settings - Fork 14
/
run_pubmed.py
178 lines (150 loc) · 7.79 KB
/
run_pubmed.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
from __future__ import division
from __future__ import print_function
import tensorflow as tf
import time
import os
from inits import *
from sampler import *
from models import GCNAdapt, GCNAdaptMix
# Set random seed
seed = 123
np.random.seed(seed)
tf.set_random_seed(seed)
# Settings
flags = tf.app.flags
FLAGS = flags.FLAGS
flags.DEFINE_string('dataset', 'cora', 'Dataset string.') # 'cora', 'citeseer', 'pubmed'
flags.DEFINE_string('model', 'gcn_adapt', 'Model string.') # 'gcn', 'gcn_appr'
flags.DEFINE_float('learning_rate', 0.001, 'Initial learning rate.')
flags.DEFINE_integer('epochs', 300, 'Number of epochs to train.')
flags.DEFINE_integer('hidden1', 16, 'Number of units in hidden layer 1.')
flags.DEFINE_float('dropout', 0.0, 'Dropout rate (1 - keep probability).')
flags.DEFINE_float('weight_decay', 5e-4, 'Weight for L2 loss on embedding matrix.')
flags.DEFINE_integer('early_stopping', 30, 'Tolerance for early stopping (# of epochs).')
flags.DEFINE_integer('max_degree', 32, 'Maximum degree for constructing the adjacent matrix.')
flags.DEFINE_string('gpu', '0', 'The gpu to be applied.')
flags.DEFINE_string('sampler_device', 'cpu', 'The device for sampling: cpu or gpu.')
flags.DEFINE_integer('rank', 128, 'The number of nodes per layer.')
flags.DEFINE_integer('skip', 0, 'If use skip connection.')
flags.DEFINE_float('var', 0.5, 'If use variance reduction.')
os.environ["CUDA_VISIBLE_DEVICES"] = str(FLAGS.gpu)
def main(rank1, rank0):
# Prepare data
adj, adj_train, adj_val_train, features, train_features, y_train, y_test, test_index = prepare_pubmed(FLAGS.dataset, FLAGS.max_degree)
print('preparation done!')
max_degree = FLAGS.max_degree
num_train = adj_train.shape[0] - 1
# num_train = adj_train.shape[0]
input_dim = features.shape[1]
scope = 'test'
if FLAGS.model == 'gcn_adapt_mix':
num_supports = 1
propagator = GCNAdaptMix
test_supports = [sparse_to_tuple(adj[test_index, :])]
test_features = [features, features[test_index, :]]
test_probs = [np.ones(adj.shape[0])]
layer_sizes = [rank1, 256]
elif FLAGS.model == 'gcn_adapt':
num_supports = 2
propagator = GCNAdapt
test_supports = [sparse_to_tuple(adj), sparse_to_tuple(adj[test_index, :])]
test_features = [features, features, features[test_index, :]]
test_probs = [np.ones(adj.shape[0]), np.ones(adj.shape[0])]
layer_sizes = [rank0, rank1, 256]
else:
raise ValueError('Invalid argument for model: ' + str(FLAGS.model))
# Define placeholders
placeholders = {
'batch': tf.placeholder(tf.int32),
'adj': tf.placeholder(tf.int32, shape=(num_train+1, max_degree)),
'adj_val': tf.placeholder(tf.float32, shape=(num_train+1, max_degree)),
'features': tf.placeholder(tf.float32, shape=train_features.shape),
'support': [tf.sparse_placeholder(tf.float32) for _ in range(num_supports)],
'prob': [tf.placeholder(tf.float32) for _ in range(num_supports)],
'features_inputs': [tf.placeholder(tf.float32, shape=(None, input_dim)) for _ in range(num_supports+1)],
'labels': tf.placeholder(tf.float32, shape=(None, y_train.shape[1])),
'labels_mask': tf.placeholder(tf.int32),
'dropout': tf.placeholder_with_default(0., shape=()),
'num_features_nonzero': tf.placeholder(tf.int32) # helper variable for sparse dropout
}
# Sampling parameters shared by the sampler and model
with tf.variable_scope(scope):
w_s = glorot([features.shape[-1], 2], name='sample_weights')
# Create sampler
if FLAGS.sampler_device == 'cpu':
with tf.device('/cpu:0'):
sampler_tf = SamplerAdapt(placeholders, input_dim=input_dim, layer_sizes=layer_sizes, scope=scope)
features_sampled, support_sampled, p_u_sampled = sampler_tf.sampling(placeholders['batch'])
else:
sampler_tf = SamplerAdapt(placeholders, input_dim=input_dim, layer_sizes=layer_sizes, scope=scope)
features_sampled, support_sampled, p_u_sampled = sampler_tf.sampling(placeholders['batch'])
# Create model
model = propagator(placeholders, input_dim=input_dim, logging=True, name=scope)
# Initialize session
config = tf.ConfigProto(device_count={"CPU": 1},
inter_op_parallelism_threads=0,
intra_op_parallelism_threads=0,
allow_soft_placement=True,
log_device_placement=False)
sess = tf.Session(config=config)
# Define model evaluation function
def evaluate(features, support, prob_norm, labels, mask, placeholders):
t_test = time.time()
feed_dict_val = construct_feed_dict_with_prob(features, support, prob_norm, labels, mask, placeholders)
outs_val = sess.run([model.loss, model.accuracy], feed_dict=feed_dict_val)
return outs_val[0], outs_val[1], (time.time() - t_test)
# Init variables
sess.run(tf.global_variables_initializer(), feed_dict={placeholders['adj']: adj_train,
placeholders['adj_val']: adj_val_train,
placeholders['features']: train_features})
# Prepare training
saver = tf.train.Saver()
save_dir = "tmp/" + FLAGS.dataset + '_' + str(FLAGS.skip) + '_' + str(FLAGS.var) + '_' + str(FLAGS.gpu)
acc_val = []
acc_train = []
train_time = []
train_time_sample = []
max_acc = 0
t = time.time()
# Train model
for epoch in range(FLAGS.epochs):
sample_time = 0
t1 = time.time()
for batch in iterate_minibatches_listinputs([y_train, np.arange(num_train)], batchsize=256, shuffle=True):
[y_train_batch, train_batch] = batch
if sum(train_batch) < 1:
continue
ts = time.time()
features_inputs, supports, probs = sess.run([features_sampled, support_sampled, p_u_sampled],
feed_dict={placeholders['batch']:train_batch})
sample_time += time.time()-ts
# Construct feed dictionary
feed_dict = construct_feed_dict_with_prob(features_inputs, supports, probs, y_train_batch, [],
placeholders)
feed_dict.update({placeholders['dropout']: FLAGS.dropout})
# Training step
outs = sess.run([model.opt_op, model.loss, model.accuracy], feed_dict=feed_dict)
acc_train.append(outs[-2])
train_time_sample.append(time.time()-t1)
train_time.append(time.time()-t1-sample_time)
# Validation
cost, acc, duration = evaluate(test_features, test_supports, test_probs, y_test, [], placeholders)
acc_val.append(acc)
# if epoch > 50 and acc>max_acc:
# max_acc = acc
# saver.save(sess, save_dir + ".ckpt")
# Print results
print("Epoch:", '%04d' % (epoch + 1), "train_loss=", "{:.5f}".format(outs[1]),
"train_acc=", "{:.5f}".format(outs[2]), "val_loss=", "{:.5f}".format(cost),
"val_acc=", "{:.5f}".format(acc), "time=", "{:.5f}".format(train_time_sample[epoch]))
train_duration = np.mean(np.array(train_time_sample))
# Testing
# if os.path.exists(save_dir + ".ckpt.index"):
# saver.restore(sess, save_dir + ".ckpt")
# print('Loaded the best ckpt.')
# test_cost, test_acc, test_duration = evaluate(test_features, test_supports, test_probs, y_test, [], placeholders)
# print("rank1 = {}".format(rank1), "rank0 = {}".format(rank0), "cost=", "{:.5f}".format(test_cost),
# "accuracy=", "{:.5f}".format(test_acc), "training time per epoch=", "{:.5f}".format(train_duration))
if __name__ == "__main__":
print("DATASET:", FLAGS.dataset)
main(FLAGS.rank,FLAGS.rank)