Try to add batch normalization in network.py, failed with dependency line.

[Franklalalala]

I used high level api, tf.layers.batch_normalization(hidden, training=True, trainable=True)
They say, you need solve the dependency like:
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
train_op = optimizer.minimize(loss)
or, it'll not learn the gamma and belta.
I tried to add these lines in trainer.py, but failed with bugs.
I see there are lines about updatate variables.

deepmd-kit/deepmd/train/trainer.py

Lines 337 to 354 in 159e45d

	def _build_training(self):
	trainable_variables = tf.trainable_variables()
	if self.run_opt.is_distrib:
	if self.scale_lr_coef > 1.:
	log.info('Scale learning rate by coef: %f', self.scale_lr_coef)
	optimizer = tf.train.AdamOptimizer(self.learning_rate*self.scale_lr_coef)
	else:
	optimizer = tf.train.AdamOptimizer(self.learning_rate)
	optimizer = self.run_opt._HVD.DistributedOptimizer(optimizer)
	else:
	optimizer = tf.train.AdamOptimizer(learning_rate = self.learning_rate)
	apply_op = optimizer.minimize(loss=self.l2_l,
	global_step=self.global_step,
	var_list=trainable_variables,
	name='train_step')
	train_ops = [apply_op] + self._extra_train_ops
	self.train_op = tf.group(*train_ops)
	log.info("built training")

Does that mean, I do not need to check the dependency anymore?
I'm new to tf, coundn't understand it.

I tried low API wrapper, always fail on line ↓
with tf.control_dependencies([assign_moving_average(pop_mean, batch_mean, decay),
assign_moving_average(pop_var, batch_var, decay)]):
return tf.identity(batch_mean), tf.identity(batch_var)
I don't know, how to slove this problem, or maybe, just leave it. ( It works without dependency line, with the high level API.)
Help!