-
Notifications
You must be signed in to change notification settings - Fork 2
/
train.py
190 lines (165 loc) · 7.75 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
185
186
187
188
189
190
from load_dataset import *
from utils import *
from model import *
import matplotlib.pyplot as plt
from sklearn.manifold import TSNE
flags.DEFINE_string('model', 'inception', 'Model string.') # gcn, gcn_cheby, inception
flags.DEFINE_float('learning_rate', 0.001, 'Initial learning rate.')
flags.DEFINE_integer('epochs', 150, 'Number of epochs to train.')
flags.DEFINE_integer('hidden1', 36, 'Number of units in hidden layer 1.')
flags.DEFINE_integer('hidden2', 18, 'Number of units in hidden layer 2.')
flags.DEFINE_integer('hidden3', 9, 'Number of units in hidden layer 3.')
flags.DEFINE_float('dropout', 0., 'Dropout rate (1 - keep probability).')
flags.DEFINE_float('weight_decay', 0., 'Weight for L2 loss on embedding matrix.')
flags.DEFINE_bool('featureless', False, 'featureless')
base_path = './data8/'
adj, num_nodes = load_adj(base_path)
labels, one_hot_labels, num_graphs, num_classes, nan_idx = load_classes2(base_path)
class_dist = [labels.tolist().count(i) / num_graphs for i in range(num_classes)]
print(class_dist)
features = load_features(base_path, is_binary=False)
features = features[~nan_idx]
class_idx = np.asarray([i for i, l in enumerate(labels) if class_dist[l] >= 0.05])
labels = labels[class_idx]
one_hot_labels = one_hot_labels[class_idx]
features = features[class_idx]
num_classes -= np.sum(np.asarray(class_dist) < 0.05).astype(int)
num_graphs = labels.shape[0]
k = (one_hot_labels != 0).any(0)
one_hot_labels = one_hot_labels[:, (one_hot_labels != 0).any(0)]
labels = np.argmax(one_hot_labels, axis=1)
new_class_dist = [labels.tolist().count(i) / num_graphs for i in range(num_classes)]
print(new_class_dist)
# sub_sample = np.arange(num_graphs)
# np.random.shuffle(sub_sample)
# num_sample = 5000
# first_idx = sub_sample[:num_sample]
# num_graphs = num_sample
# labels = labels[first_idx]
# one_hot_labels = one_hot_labels[first_idx]
# features = features[first_idx]
# num_train = num_graphs
train_proportion = 0.7
num_train = int(num_graphs * train_proportion)
idx = np.arange(num_graphs)
np.random.shuffle(idx)
# indices of train and test
train_idx = idx[:num_train]
test_idx = idx[num_train:]
# collecting train samples
train_labels = labels[train_idx]
train_one_hot_labels = one_hot_labels[train_idx]
train_features = features[train_idx]
# collecting test samples
test_labels = labels[test_idx]
test_one_hot_labels = one_hot_labels[test_idx]
test_features = features[test_idx]
train_sparse_features = []
test_sparse_features = []
sparse_features = []
train_graph_weights = [1 / class_dist[train_labels[i]] for i in range(num_train)]
for i in range(num_graphs):
sparse_features.append(sparse_to_tuple(sp.coo_matrix(np.expand_dims(np.transpose(features[i, :]), 1))))
for i in range(num_train):
train_sparse_features.append(sparse_to_tuple(sp.coo_matrix(np.expand_dims(np.transpose(train_features[i, :]), 1))))
for i in range(num_graphs - num_train):
test_sparse_features.append(sparse_to_tuple(sp.coo_matrix(np.expand_dims(np.transpose(test_features[i, :]), 1))))
if FLAGS.model == 'gcn_cheby':
locality1 = 8
locality2 = 7
locality3 = 6
locality = [locality1, locality2, locality3] # locality sizes of different blocks
num_supports = np.max(locality) + 1
support = chebyshev_polynomials(adj, num_supports - 1)
elif FLAGS.model == 'inception':
locality_sizes = [7, 5, 3]
num_supports = np.max(locality_sizes) + 1
support = chebyshev_polynomials(adj, num_supports - 1)
elif FLAGS.model == 'gcn':
num_supports = 1
support = [preprocess_adj(adj)]
else:
raise NotImplementedError
placeholders = {
'support': [tf.sparse_placeholder(tf.float32, name='support_{}'.format(i)) for i in range(num_supports)],
'features': tf.sparse_placeholder(tf.float32),
'labels': tf.placeholder(tf.float32, shape=(one_hot_labels.shape[1])),
'dropout': tf.placeholder_with_default(0., shape=()),
'weight': tf.placeholder(tf.float32),
'num_features_nonzero': tf.placeholder(tf.int32) # helper variable for sparse dropout
}
# model definition
if FLAGS.model == 'gcn_cheby':
model = CheybyGCN(placeholders, input_dim=1, num_class=num_classes, locality=locality)
elif FLAGS.model == 'inception':
model = InceptionGCN(placeholders, input_dim=1, num_class=num_classes,
locality_sizes=locality_sizes, is_pool=True)
else:
model = SimpleGCN(placeholders, input_dim=1, num_class=num_classes)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
test_result = []
for epoch in range(FLAGS.epochs):
# print('Starting epoch {}'.format(epoch + 1))
cnt = 0
sum_loss = 0
train_acc_classes = np.zeros((num_classes, num_classes), dtype=np.int32)
test_acc_classes = np.zeros((num_classes, num_classes), dtype=np.int32)
for i in range(num_train):
train_feed_dict = construct_feed_dict(train_sparse_features[i], support, train_one_hot_labels[i],
train_graph_weights[i], placeholders)
train_feed_dict.update({placeholders['dropout']: FLAGS.dropout})
_, loss, acc, out = sess.run([model.opt_op, model.loss, model.accuracy, model.outputs],
feed_dict=train_feed_dict)
train_acc_classes[train_labels[i], np.argmax(out, 1)[0]] += 1
# print('Graph {}: '.format(i + 1), 'Loss={}, '.format(loss), 'Acc={}'.format(acc))
# train_acc_classes[train_labels[i], prediction] += 1
cnt += acc
sum_loss += loss
print('Epoch {}:'.format(epoch + 1), 'acc={:.4f}, loss={:.4f}'.format(cnt / float(num_train),
sum_loss / float(num_train)))
cnt = 0
for i in range(num_graphs - num_train):
test_feed_dict = construct_feed_dict(test_sparse_features[i], support, test_one_hot_labels[i], 1,
placeholders)
test_feed_dict.update({placeholders['dropout']: 0.})
acc, out = sess.run([model.accuracy, model.outputs], feed_dict=test_feed_dict)
test_acc_classes[test_labels[i], np.argmax(out, 1)[0]] += 1
# test_acc_classes[test_labels[i], prediction] += 1
cnt += acc
test_acc = cnt / float(num_graphs - num_train)
test_result.append(test_acc)
print('Test accuracy: {:.4f}'.format(test_acc))
print('train confusion matrix: \n', train_acc_classes)
print('test confusion matrix: \n', test_acc_classes)
print("Optimization finished!")
plt.plot(test_result)
plt.show()
print('Storing graph embedding')
embedding_level = 4
with open('./embedding/graph_embedding21.csv', 'w') as csv_file:
writer = csv.writer(csv_file)
header = ['id']
for i in range(FLAGS.hidden3):
header.append('emb_{}'.format(i))
writer.writerow(header)
embeddings = []
for i in range(num_graphs):
feed_dict = construct_feed_dict(sparse_features[i], support, one_hot_labels[i], 1, placeholders)
feed_dict.update({placeholders['dropout']: 0.})
embedding = sess.run(model.activations[embedding_level], feed_dict=feed_dict)
row = [i + 1]
for item in embedding.tolist()[0]:
row.append(item)
writer.writerow(row)
embeddings.append(embedding.tolist()[0])
# print('Plotting t-SNE')
# embeddings = np.asarray(embeddings)
# reduced_embedding = TSNE(n_components=2).fit_transform(embeddings)
# color_names = ['b', 'g', 'r', 'y']
# colors = [color_names[label] for label in labels]
# num_samples = 5000
# plt.scatter(reduced_embedding[:num_samples, 0], reduced_embedding[:num_samples, 1],
# marker='.',
# c=colors[:num_samples])
# plt.show()