Any gus succeed to update moving_mean and moving_variance? about maml HOT 1 CLOSED

I found out the solution.
Put update_ops in meta_task function to make update_ops in the scope of while_loop(refer to： tensorflow/tensorflow#6087).
and set training=your_train_or_test_flag for tf.layers.batch_normallization function.
Do NOT forget to use tf.train.Saver(tf.global_variables()) to save moving_mean, moving_variance, otherwise the moving_mean, moving_variance will not be saved into checkpoint.

def meta_task():
...
			update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
			with tf.control_dependencies(update_ops):
				support_loss = tf.identity(support_loss)

from maml.