def train(self):
"""
Method for:
1. training the embedding.
2. logging.
This method is inherited by GEMSEC and DeepWalk variants without an override.
"""
self.current_step = 0
self.current_gamma = self.args.initial_gamma
self.log = log_setup(self.args)
with tf.Session(graph=self.computation_graph) as session:
self.init.run()
print("Model Initialized.")
for repetition in range(self.args.num_of_walks):
random.shuffle(self.nodes)
self.optimization_time = 0
self.average_loss = 0
epoch_printer(repetition)
for node in tqdm(self.nodes):
self.current_step = self.current_step + 1
self.current_gamma = gamma_incrementer(
self.current_step, self.args.initial_gamma,
self.args.final_gamma, self.current_gamma,
self.true_step_size)
feed_dict = self.feed_dict_generator(
self.walks[self.current_step - 1], self.current_step,
self.current_gamma)
start = time.time()
_, loss = session.run([self.train_op, self.loss],
feed_dict=feed_dict)
end = time.time()
self.optimization_time = self.optimization_time + (end -
start)
self.average_loss = self.average_loss + loss
print("")
self.average_loss = self.average_loss / self.vocab_size
self.final_embeddings = self.walker_layer.embedding_matrix.eval(
)
if "GEMSEC" in self.args.model:
self.c_means = self.cluster_layer.cluster_means.eval()
self.modularity_score, assignments = neural_modularity_calculator(
self.graph, self.final_embeddings, self.c_means)
else:
self.modularity_score, assignments = classical_modularity_calculator(
self.graph, self.final_embeddings, self.args)
self.log = log_updater(self.log, repetition, self.average_loss,
self.optimization_time,
self.modularity_score)
tab_printer(self.log)
if "GEMSEC" in self.args.model:
initiate_dump_gemsec(self.log, assignments, self.args,
self.final_embeddings, self.c_means)
else:
initiate_dump_dw(self.log, assignments, self.args,
self.final_embeddings)
def train(self):
"""
Method for training the embedding, logging.
"""
self.current_step = 0
self.log = log_setup(self.args)
with tf.Session(graph=self.computation_graph) as session:
self.init.run()
print("Model Initialized.")
for repetition in range(0, self.args.epochs):
random.shuffle(self.nodes)
self.optimization_time = 0
self.average_loss = 0
epoch_printer(repetition)
for i in tqdm(range(0,
len(self.nodes) / self.args.batch_size)):
self.current_step = self.current_step + 1
feed_dict = self.feed_dict_generator(
self.nodes[i * self.args.batch_size:(i + 1) *
self.args.batch_size], self.current_step)
start = time.time()
_, loss = session.run([self.train_op, self.loss],
feed_dict=feed_dict)
end = time.time()
self.optimization_time = self.optimization_time + (end -
start)
self.average_loss = self.average_loss + loss
print("")
self.average_loss = (self.average_loss / i)
self.log = log_updater(self.log, repetition, self.average_loss,
self.optimization_time)
tab_printer(self.log)
self.features = self.factorization_layer.embedding_node.eval()
data_saver(self.features, self.args.embedding_output)