def _get_dense_feature(self, nodes, feature_ids):
f = tf_euler.get_dense_feature(
nodes,
feature_ids,
[1],
)
return self.session.run(f[0]).tolist()
def get_graph_label(self, sample_graph):
graph_idx = sample_graph.indices[:, 1]
graph_mask = tf.equal(graph_idx, 0)
graph_mask = tf.cast(graph_mask, dtype=tf.int32)
graph_mask = tf.reshape(graph_mask, [-1])
graph_node = tf.boolean_mask(sample_graph.values, graph_mask)
graph_label, = tf_euler.get_dense_feature(graph_node,
self.params['label'], [1])
graph_label = tf.reshape(tf.cast(graph_label, dtype=tf.int32), [-1])
graph_label = tf.one_hot(graph_label, self.params['num_classes'])
return graph_label
def __call__(self, inputs):
label, = tf_euler.get_dense_feature(inputs, [self.label_idx],
[self.label_dim])
embedding = self.embed(inputs)
logit = self.out_fc(embedding)
metric = self.metric_class(label, tf.nn.sigmoid(logit))
loss = tf.nn.sigmoid_cross_entropy_with_logits(labels=label,
logits=logit)
loss = tf.reduce_mean(loss)
#add logit output
return (embedding, loss, self.metric_name, metric, logit)
def call(self, inputs):
samples = tf_euler.sample_fanout(inputs, self.metapath, self.fanouts)[0]
hidden = [
tf_euler.get_dense_feature(sample,
[self.feature_idx], [self.feature_dim])[0]
for sample in samples]
for layer in range(self.num_layers):
aggregator = self.aggregators[layer]
next_hidden = []
for hop in range(self.num_layers - layer):
neigh_shape = [-1, self.fanouts[hop], self.dims[layer]]
h = aggregator((hidden[hop], tf.reshape(hidden[hop + 1], neigh_shape)))
next_hidden.append(h)
hidden = next_hidden
return hidden[0]
def sampler(self, inputs):
labels = tf_euler.get_dense_feature(inputs, [self.label_idx],
[self.label_dim])[0]
return inputs, labels
def __call__(self, inputs):
label, = tf_euler.get_dense_feature(inputs, [self.label_idx],
[self.label_dim])
return label
def to_x(self, n_id):
x, = tf_euler.get_dense_feature(n_id, self.feature_idx,
self.feature_dim)
return x
