Tensorflow `tf.layers.batch_normalization` 不会向 `tf.GraphKeys.UPDATE_OPS` 添加更新操作

以下代码（复制/粘贴可运行）说明了如何使用tf.layers.batch_normalization.

import tensorflow as tf
bn = tf.layers.batch_normalization(tf.constant([0.0]))
print(tf.get_collection(tf.GraphKeys.UPDATE_OPS))

> []     # UPDATE_OPS collection is empty

使用 TF 1.5，文档（下面引用）明确指出UPDATE_OPS 不应为空在这种情况下 （https://www.tensorflow.org/api_docs/python/tf/layers/batch_normalization https://www.tensorflow.org/api_docs/python/tf/layers/batch_normalization):

注意：训练时，moving_mean和moving_variance需要 更新。默认情况下，更新操作位于tf.GraphKeys.UPDATE_OPS，因此需要将它们添加为依赖项 train_op。例如：

update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(update_ops):
    train_op = optimizer.minimize(loss)

只需将您的代码更改为训练模式（通过设置training标记为True）如中提到的quote https://www.tensorflow.org/api_docs/python/tf/layers/batch_normalization:

注：当training， moving_mean 和 moving_variance 需要更新。默认情况下，更新操作放置在 tf.GraphKeys.UPDATE_OPS 中，因此需要将它们添加为 train_op 的依赖项。

 import tensorflow as tf
 bn = tf.layers.batch_normalization(tf.constant([0.0]), training=True)
 print(tf.get_collection(tf.GraphKeys.UPDATE_OPS))

将输出：

[< tf.Tensor 'batch_normalization/AssignMovingAvg:0' shape=(1,) dtype=float32_ref>, 
 < tf.Tensor 'batch_normalization/AssignMovingAvg_1:0' shape=(1,) dtype=float32_ref>]

Gamma 和 Beta 最终出现在 TRAINABLE_VARIABLES 集合中：

print(tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES))

[<tf.Variable 'batch_normalization/gamma:0' shape=(1,) dtype=float32_ref>, 
 <tf.Variable 'batch_normalization/beta:0' shape=(1,) dtype=float32_ref>]