def build(self):
if self.unsupervised:
self.ego_flow = gl.EgoFlow(
self._sample_seed,
self._positive_sample,
self._receptive_fn,
self.src_ego_spec,
dst_ego_spec=self.src_ego_spec, # h**o graph.
negative_sample=self._negative_sample,
full_graph_mode=self.full_graph_mode)
self.pos_src_ego_tensor = self.ego_flow.pos_src_ego_tensor
self.pos_dst_ego_tensor = self.ego_flow.pos_dst_ego_tensor
self.neg_dst_ego_tensor = self.ego_flow.neg_dst_ego_tensor
self.iterator = self.ego_flow.iterator
pos_src_emb = self.encoders['src'].encode(self.pos_src_ego_tensor)
pos_dst_emb = self.encoders['dst'].encode(self.pos_dst_ego_tensor)
neg_dst_emb = self.encoders['dst'].encode(self.neg_dst_ego_tensor)
self.loss = self._unsupervised_loss(pos_src_emb, pos_dst_emb,
neg_dst_emb)
self.loss = self.loss[0]
else:
self.ego_flow = gl.EgoFlow(self._sample_seed,
self._positive_sample,
self._receptive_fn,
self.src_ego_spec,
full_graph_mode=self.full_graph_mode)
self.iterator = self.ego_flow.iterator
self.pos_src_ego_tensor = self.ego_flow.pos_src_ego_tensor
src_emb = self.encoders['src'].encode(self.pos_src_ego_tensor)
labels = self.pos_src_ego_tensor.src.labels
self.loss = self._supervised_loss(src_emb, labels)
return self.loss, self.iterator
def build(self):
ego_flow = gl.EgoFlow(self._sample_seed,
self._positive_sample,
self._receptive_fn,
self.node_ego_spec,
dst_ego_spec=self.node_ego_spec,
edge_ego_spec=self.edge_ego_spec,
negative_sample=self._negative_sample)
iterator = ego_flow.iterator
pos_src_ego_tensor = ego_flow.pos_src_ego_tensor
pos_dst_ego_tensor = ego_flow.pos_dst_ego_tensor
neg_src_ego_tensor = ego_flow.neg_src_ego_tensor
neg_dst_ego_tensor = ego_flow.neg_dst_ego_tensor
pos_edge_ego_tensor = ego_flow.pos_edge_ego_tensor
neg_edge_ego_tensor = ego_flow.neg_edge_ego_tensor
pos_src_emb = self.encoders['src'].encode(pos_src_ego_tensor)
pos_dst_emb = self.encoders['dst'].encode(pos_dst_ego_tensor)
neg_src_emb = self.encoders['src'].encode(neg_src_ego_tensor)
neg_dst_emb = self.encoders['dst'].encode(neg_dst_ego_tensor)
pos_edge_emb = self.encoders['edge'].encode(pos_edge_ego_tensor)
neg_edge_emb = self.encoders['edge'].encode(neg_edge_ego_tensor)
loss = self._unsupervised_loss_with_edge(pos_src_emb, pos_edge_emb,
neg_src_emb, pos_dst_emb,
neg_edge_emb, neg_dst_emb)
return loss[0], iterator
def node_embedding(self, type):
node_ego_flow = gl.EgoFlow(self._save_node_sample_seed,
self._positive_sample, self._receptive_fn,
self.ego_spec)
iterator = node_ego_flow.iterator
ego_tensor = node_ego_flow.pos_src_ego_tensor
src_emb = self.encoders['src'].encode(ego_tensor)
src_ids = ego_tensor.src.ids
return src_ids, src_emb, iterator
def test_acc(self):
test_ego_flow = gl.EgoFlow(self._test_sample_seed,
self._positive_sample,
self._receptive_fn,
self.src_ego_spec,
full_graph_mode=self.full_graph_mode)
test_iterator = test_ego_flow.iterator
test_pos_src_ego_tensor = test_ego_flow.pos_src_ego_tensor
test_logits = self.encoders['src'].encode(test_pos_src_ego_tensor)
test_labels = test_pos_src_ego_tensor.src.labels
return self._accuracy(test_logits, test_labels), test_iterator
def val_acc(self):
val_ego_flow = gl.EgoFlow(self._val_sample_seed,
self._positive_sample,
self._receptive_fn,
self.src_ego_spec,
full_graph_mode=self.full_graph_mode)
val_iterator = val_ego_flow.iterator
val_pos_src_ego_tensor = val_ego_flow.pos_src_ego_tensor
val_logits = self.encoders['src'].encode(val_pos_src_ego_tensor)
val_labels = val_pos_src_ego_tensor.src.labels
return self._accuracy(val_logits, val_labels), val_iterator
def node_embedding(self, type='u'):
"""Return embeddings for saving"""
if type == 'u':
# for save u embeddings
ego_flow = gl.EgoFlow(self._u_sample_seed, self._positive_sample,
self._receptive_fn, self.u_ego_spec)
iterator = ego_flow.iterator
ego_tensor = ego_flow.pos_src_ego_tensor
emb = self.encoders['src'].encode(ego_tensor)
ids = ego_tensor.src.ids
else:
# for save i embeddings
ego_flow = gl.EgoFlow(self._i_sample_seed, self._positive_sample,
self._receptive_fn, self.i_ego_spec)
iterator = ego_flow.iterator
ego_tensor = ego_flow.pos_src_ego_tensor
emb = self.encoders['dst'].encode(ego_tensor)
ids = ego_tensor.src.ids
return ids, emb, iterator
def node_embedding(self, type='entity'):
"""Return embeddings for saving"""
if type == 'entity':
# for save entity embeddings
ego_flow = gl.EgoFlow(self._entity_sample_seed,
self._positive_sample, self._receptive_fn,
self.node_ego_spec)
iterator = ego_flow.iterator
ego_tensor = ego_flow.pos_src_ego_tensor
emb = self.encoders['src'].encode(ego_tensor)
ids = ego_tensor.src.ids
else:
# for save relation embeddings
ego_flow = gl.EgoFlow(self._relation_sample_seed,
self._positive_sample, self._receptive_fn,
self.node_ego_spec)
iterator = ego_flow.iterator
ego_tensor = ego_flow.pos_src_ego_tensor
emb = self.encoders['edge'].encode(ego_tensor)
ids = ego_tensor.src.ids
return ids, emb, iterator
def build(self):
ego_flow = gl.EgoFlow(self._sample_seed,
self._positive_sample,
self._receptive_fn,
self.src_ego_spec,
full_graph_mode=self.full_graph_mode)
iterator = ego_flow.iterator
pos_src_ego_tensor = ego_flow.pos_src_ego_tensor
src_emb = self.encoders['src'].encode(pos_src_ego_tensor)
labels = pos_src_ego_tensor.src.labels
loss = self._supervised_loss(src_emb, labels)
return loss, iterator
def build(self):
self.ego_flow = gl.EgoFlow(self._sample_seed,
self._positive_sample,
self._receptive_fn,
self.ego_spec,
dst_ego_spec=self.ego_spec, # h**o graph.
negative_sample=None) # use sampled softmax loss.
self.iterator = self.ego_flow.iterator
self.pos_src_ego_tensor = self.ego_flow.pos_src_ego_tensor
self.pos_dst_ego_tensor = self.ego_flow.pos_dst_ego_tensor
pos_src_emb = self.encoders['src'].encode(self.pos_src_ego_tensor)
pos_dst_emb = self.pos_dst_ego_tensor.src.ids # use sampled softmax
self.loss = self._unsupervised_loss(pos_src_emb, pos_dst_emb)
self.loss = self.loss[0]
return self.loss, self.iterator
def build(self):
ego_flow = gl.EgoFlow(
self._sample_seed,
self._positive_sample,
self._receptive_fn,
self.ego_spec,
dst_ego_spec=self.ego_spec, # h**o graph.
negative_sample=self._negative_sample)
iterator = ego_flow.iterator
pos_src_ego_tensor = ego_flow.pos_src_ego_tensor
pos_dst_ego_tensor = ego_flow.pos_dst_ego_tensor
neg_dst_ego_tensor = ego_flow.neg_dst_ego_tensor
pos_src_emb = self.encoders['src'].encode(pos_src_ego_tensor)
pos_dst_emb = self.encoders['dst'].encode(pos_dst_ego_tensor)
neg_dst_emb = self.encoders['dst'].encode(neg_dst_ego_tensor)
loss = self._unsupervised_loss(pos_src_emb, pos_dst_emb, neg_dst_emb)
loss = loss[0]
return loss, iterator
