def get_neighbors(self, n_id):
neighbors = []
neighbor_src = []
total_fanout = 0
for i in range(len(self.metapath)):
n_id = tf.reshape(n_id, [-1])
if i == len(self.metapath) - 1:
one_neighbor = tf_euler.get_full_neighbor(
n_id, self.metapath[i])[0]
else:
total_fanout += self.fanouts[i]
one_neighbor = tf_euler.sample_node(total_fanout,
self.metapath[i][0])
one_neighbor, _ = tf.unique(one_neighbor)
one_neighbor = tf_euler.sparse_get_adj(n_id, one_neighbor,
self.metapath[i], -1,
-1)
neighbors.append(tf.reshape(one_neighbor.values, [-1]))
one_indices = one_neighbor.indices[:, 0]
neighbor_src.append(tf.cast(one_indices, tf.int32))
new_n_id = tf.reshape(one_neighbor.values, [-1])
n_id = tf.concat([new_n_id, n_id], axis=0)
n_id, _ = tf.unique(n_id)
return neighbors, neighbor_src
def sampler(self, inputs):
batch_size = tf.size(inputs)
num_negs = self.num_negs
src, pos, negs = [], [], []
for i in range(self.walk_num):
path = tf_euler.random_walk(inputs,
[self.edge_type] * self.walk_len,
default_node=self.max_id + 1)
pair = tf_euler.gen_pair(path, self.left_win_size,
self.right_win_size)
num_pairs = pair.shape[1]
src_, pos_ = tf.split(pair, [1, 1], axis=-1)
negs_ = tf_euler.sample_node(batch_size * num_pairs * num_negs,
self.node_type)
src_ = tf.reshape(src_, [batch_size * num_pairs, 1])
pos_ = tf.reshape(pos_, [batch_size * num_pairs, 1])
negs_ = tf.reshape(negs_, [batch_size * num_pairs, num_negs])
src.append(src_)
pos.append(pos_)
negs.append(negs_)
if self.walk_num == 1:
return src[0], pos[0], negs[0] # no need to concat
src = tf.concat(src, axis=0, name='src')
pos = tf.concat(pos, axis=0, name='pos')
negs = tf.concat(negs, axis=0, name='negs')
return src, pos, negs
def to_sample(self, inputs):
batch_size = tf.size(inputs)
src = tf.expand_dims(inputs, -1)
pos = tf_euler.sample_neighbor(inputs, self.edge_type, 1,
self.max_id + 1)[0]
negs = tf_euler.sample_node(batch_size * self.num_negs, self.node_type)
negs = tf.reshape(negs, [batch_size, self.num_negs])
return src, pos, negs
def __call__(self, inputs):
batch_size = tf.shape(inputs)[0]
neg_nodes_group = []
for n_type in self.neg_type:
neg_node = tf_euler.sample_node(batch_size * self.num_negs, n_type)
neg_node = tf.reshape(neg_node, [batch_size, self.num_negs])
neg_nodes_group.append(neg_node)
if len(self.neg_type) == 1:
return neg_nodes_group[0]
else:
return neg_nodes_group
def sampler(self, inputs):
batch_size = tf.size(inputs)
path = tf_euler.random_walk(inputs, [self.edge_type] * self.walk_len,
default_node=self.max_id + 1)
pair = tf_euler.gen_pair(path, self.left_win_size, self.right_win_size)
num_pairs = pair.shape[1]
src, pos = tf.split(pair, [1, 1], axis=-1)
negs = tf_euler.sample_node(batch_size * num_pairs * self.num_negs,
self.node_type)
src = tf.reshape(src, [batch_size * num_pairs, 1])
pos = tf.reshape(pos, [batch_size * num_pairs, 1])
negs = tf.reshape(negs, [batch_size * num_pairs, self.num_negs])
return src, pos, negs
def sample_subgraph_fanout(self, batch_size, degrees, edge_types):
tf.reset_default_graph()
self.session = tf.Session()
# sample source nodes at random
src_nodes = tf_euler.sample_node(batch_size, tf_euler.ALL_NODE_TYPE)
src_nodes = self.session.run(src_nodes)
batchs = []
for src_node in src_nodes:
cur_nodes = np.asarray([src_node]).astype(np.int64)
neighbors, weights, etypes = self.sample_fanout(
cur_nodes, edge_types, degrees)
sample = [[neighbors[0][0]]]
for s in zip(neighbors[1:], weights, etypes):
sample.append([i.tolist() for i in s])
batchs.append(sample)
return batchs
labels = tf_euler.get_dense_feature(inputs, [self.label_idx],
[self.label_dim])[0]
return inputs, labels
def decoder(self, embedding, labels):
logits = self.predict_layer(embedding)
loss = tf.nn.softmax_cross_entropy_with_logits(labels=labels,
logits=logits)
#predictions = tf.floor(tf.nn.sigmoid(logits) + 0.5)
#f1 = tf_euler.metrics.f1_score(labels, predictions)
return tf.reduce_mean(loss) #, f1
tf_euler.initialize_embedded_graph('reddit')
source = tf_euler.sample_node(1000, 0)
source.set_shape([1000])
model = GraphSage(0, 41, [[0], [0]], [4, 4], 64, 1, 602)
_, loss = model(source)
global_step = tf.train.get_or_create_global_step()
train_op = tf.train.AdamOptimizer(0.03).minimize(loss, global_step)
tf.logging.set_verbosity(tf.logging.INFO)
with tf.train.MonitoredTrainingSession(hooks=[
tf.train.LoggingTensorHook({
'step': global_step,
'loss': loss
}, 100),
tf.train.StopAtStepHook(2000)
[self.label_dim])[0]
return inputs, labels
def decoder(self, embedding, labels):
logits = self.predict_layer(embedding)
loss = tf.nn.sigmoid_cross_entropy_with_logits(labels=labels,
logits=logits)
predictions = tf.floor(tf.nn.sigmoid(logits) + 0.5)
f1 = tf_euler.metrics.f1_score(labels, predictions)
return tf.reduce_mean(loss), f1
if __name__ == '__main__':
print("begin....")
tf_euler.initialize_embedded_graph('ppi') # 图数据目录
source = tf_euler.sample_node(512, 0)
source.set_shape([512])
model = GraphSage(0, 121, [[0], [0]], [10, 10], 256, 1, 50)
_, loss, metric_name, metric = model(source)
global_step = tf.train.get_or_create_global_step()
train_op = tf.train.AdamOptimizer(0.01).minimize(loss, global_step)
tf.logging.set_verbosity(tf.logging.INFO)
with tf.train.MonitoredTrainingSession(hooks=[
tf.train.LoggingTensorHook(
{
'step': global_step,
'loss': loss,
metric_name: metric
def get_train_from_input(self, inputs, params):
result = tf_euler.sample_node(inputs, params['train_node_type'])
result.set_shape([self.params['batch_size']])
return result
def generate_negative(self, batch_size):
return tf_euler.sample_node(batch_size * self.num_negs, self.node_type)
def decoder(self, embedding, embedding_pos, embedding_negs):
logits = tf.matmul(embedding, embedding_pos, transpose_b=True)
neg_logits = tf.matmul(embedding, embedding_negs, transpose_b=True)
true_xent = tf.nn.sigmoid_cross_entropy_with_logits(
labels=tf.ones_like(logits), logits=logits)
negative_xent = tf.nn.sigmoid_cross_entropy_with_logits(
labels=tf.zeros_like(neg_logits), logits=neg_logits)
loss = tf.reduce_sum(true_xent) + tf.reduce_sum(negative_xent)
mrr = tf_euler.metrics.mrr_score(logits, neg_logits)
return loss, mrr
if __name__ == '__main__':
print("begin....")
tf_euler.initialize_embedded_graph('ppi') # 图数据目录
source = tf_euler.sample_node(128, tf_euler.ALL_NODE_TYPE)
source.set_shape([128])
model = DeepWalk(tf_euler.ALL_NODE_TYPE, [0, 1], 56944, 256)
_, loss, metric_name, metric = model(source)
global_step = tf.train.get_or_create_global_step()
train_op = tf.train.GradientDescentOptimizer(0.2).minimize(
loss, global_step)
tf.logging.set_verbosity(tf.logging.INFO)
with tf.train.MonitoredTrainingSession(hooks=[
tf.train.LoggingTensorHook(
{
'step': global_step,
'loss': loss,
