-
Notifications
You must be signed in to change notification settings - Fork 91
/
demo_distributed_mean_pool.py
161 lines (113 loc) · 5.3 KB
/
demo_distributed_mean_pool.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
# coding=utf-8
import os
# multi-gpu ids
os.environ["CUDA_VISIBLE_DEVICES"] = "0,1"
import tf_geometric as tfg
import tensorflow as tf
import numpy as np
from sklearn.model_selection import train_test_split
batch_size = 1024
drop_rate = 0.4
# TU Datasets: https://ls11-www.cs.tu-dortmund.de/staff/morris/graphkerneldatasets
graph_dicts = tfg.datasets.TUDataset("NCI1").load_data()
# Since a TU dataset may contain node_labels, node_attributes etc., each of which can be used as node features
# We process each graph as a dict and return a list of dict for graphs
# You can easily construct you Graph object with the data dict
num_node_labels = np.max([np.max(graph_dict["node_labels"]) for graph_dict in graph_dicts]) + 1
def convert_node_labels_to_one_hot(node_labels):
num_nodes = len(node_labels)
x = np.zeros([num_nodes, num_node_labels], dtype=np.float32)
x[list(range(num_nodes)), node_labels] = 1.0
return x
def construct_graph(graph_dict):
return tfg.Graph(
x=convert_node_labels_to_one_hot(graph_dict["node_labels"]),
edge_index=graph_dict["edge_index"],
y=graph_dict["graph_label"] # graph_dict["graph_label"] is a list with one int element
)
graphs = [construct_graph(graph_dict) for graph_dict in graph_dicts]
num_classes = np.max([graph.y[0] for graph in graphs]) + 1
train_graphs, test_graphs = train_test_split(graphs, test_size=0.1)
def create_graph_generator(graphs, batch_size, infinite=False, shuffle=False):
while True:
dataset = tf.data.Dataset.range(len(graphs))
if shuffle:
dataset = dataset.shuffle(2000)
dataset = dataset.batch(batch_size)
for batch_graph_index in dataset:
batch_graph_list = [graphs[i] for i in batch_graph_index]
batch_graph = tfg.BatchGraph.from_graphs(batch_graph_list)
yield (batch_graph.x, batch_graph.edge_index, batch_graph.node_graph_index), batch_graph.y
if not infinite:
break
class MeanPoolNetwork(tf.keras.Model):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.gcn0 = tfg.layers.GCN(256, activation=tf.nn.relu)
self.gcn1 = tfg.layers.GCN(256, activation=tf.nn.relu)
self.dropout = tf.keras.layers.Dropout(drop_rate)
self.dense = tf.keras.layers.Dense(num_classes)
def call(self, inputs, training=None, mask=None):
x, edge_index, node_graph_index = inputs
# bug fix for distributed training
node_graph_index = tf.reshape(node_graph_index, [-1])
# GCN Encoder
h = self.gcn0([x, edge_index], training=training)
h = self.dropout(h, training=training)
h = self.gcn1([h, edge_index], training=training)
# Mean Pooling
h = tfg.nn.mean_pool(h, node_graph_index)
h = self.dropout(h, training=training)
# Predict Graph Labels
h = self.dense(h)
return h
strategy = tf.distribute.MirroredStrategy()
def train_dataset_fn(ctx):
def create_replica_train_generator():
return create_graph_generator(train_graphs, batch_size // strategy.num_replicas_in_sync, infinite=True,
shuffle=True)
return tf.data.Dataset.from_generator(
create_replica_train_generator,
output_types=((tf.float32, tf.int32, tf.int32), tf.int32),
output_shapes=(
(tf.TensorShape([None, graphs[0].x.shape[1]]), tf.TensorShape([2, None]), tf.TensorShape([None])),
tf.TensorShape([None])
)
)
distributed_train_dataset = strategy.experimental_distribute_datasets_from_function(train_dataset_fn)
# The model will automatically use all seen GPUs defined by "CUDA_VISIBLE_DEVICES" for distributed training
with strategy.scope():
model = MeanPoolNetwork()
def cross_entropy(y_true, logits):
y_true = tf.cast(y_true, tf.int32)
losses = tf.nn.softmax_cross_entropy_with_logits(
logits=logits,
labels=tf.one_hot(y_true, depth=num_classes)
)
return tf.reduce_mean(losses)
model.compile(
optimizer=tf.keras.optimizers.Adam(learning_rate=5e-3),
loss=cross_entropy
)
def forward(batch_graph, training=False):
return model([batch_graph.x, batch_graph.edge_index, batch_graph.node_graph_index], training=training)
def evaluate():
corrects = []
for (x, edge_index, node_graph_index), y in create_graph_generator(test_graphs, batch_size, shuffle=False,
infinite=False):
logits = model([x, edge_index, node_graph_index])
preds = tf.argmax(logits, axis=-1)
corrects.append(tf.equal(preds, y))
corrects = tf.concat(corrects, axis=0)
accuracy = tf.reduce_mean(tf.cast(corrects, tf.float32))
return accuracy
class EvaluationCallback(tf.keras.callbacks.Callback):
def on_epoch_end(self, epoch, logs=None):
if epoch % 10 == 0:
test_accuracy = evaluate()
print("\nepoch = {}\ttest_accuracy = {}".format(epoch, test_accuracy))
# The model will automatically use all seen GPUs defined by "CUDA_VISIBLE_DEVICES" for distributed training
model.fit(distributed_train_dataset, steps_per_epoch=len(graphs) // batch_size, epochs=201,
callbacks=[EvaluationCallback()], verbose=1)
accuracy = evaluate()
print("\nfinal test_accuracy = {}".format(accuracy))