tf.GraphKeys

tf.GraphKeys
   用于存放许多图表集合的标准名称（`GraphKeys`类包含许多集合的标准名称）。

   标准库使用各种众所周知的名称来收集和恢复与图表相关联的变量（使用tf.add_to_collection收集、使用tf.get_collection恢复）。 例如，`tf.Optimizer`子类默认优化在tf.GraphKeys.TRAINABLE_VARIABLES下收集的变量（如果没有指定），但也可以传递显式的变量列表。

官方文档：
class GraphKeys(object):

   """Standard names to use for graph collections.

  The standard library uses various well-known names to collect and retrieve values associated with a graph. For example, the `tf.Optimizer` subclasses default to optimizing the variables collected under tf.GraphKeys.TRAINABLE_VARIABLES` if none is specified, but it is also possible to pass an explicit list of variables.
   

  The following standard keys are defined:

  * `GLOBAL_VARIABLES`: the default collection of `Variable` objects, shared
    across distributed environment (model variables are subset of these). See
    @{tf.global_variables}
    for more details.
    Commonly, all `TRAINABLE_VARIABLES` variables will be in `MODEL_VARIABLES`,
    and all `MODEL_VARIABLES` variables will be in `GLOBAL_VARIABLES`.
  * `LOCAL_VARIABLES`: the subset of `Variable` objects that are local to each
    machine. Usually used for temporarily variables, like counters.
    Note: use `tf.contrib.framework.local_variable` to add to this collection.
  * `MODEL_VARIABLES`: the subset of `Variable` objects that are used in the
    model for inference (feed forward). Note: use
    `tf.contrib.framework.model_variable` to add to this collection.
  * `TRAINABLE_VARIABLES`: the subset of `Variable` objects that will
    be trained by an optimizer. See
    @{tf.trainable_variables}
    for more details.
  * `SUMMARIES`: the summary `Tensor` objects that have been created in the
    graph. See
    @{tf.summary.merge_all}
    for more details.
  * `QUEUE_RUNNERS`: the `QueueRunner` objects that are used to
    produce input for a computation. See
    @{tf.train.start_queue_runners}
    for more details.
  * `MOVING_AVERAGE_VARIABLES`: the subset of `Variable` objects that will also
    keep moving averages.  See
    @{tf.moving_average_variables}
    for more details.
  * `REGULARIZATION_LOSSES`: regularization losses collected during graph
    construction.

  The following standard keys are _defined_, but their collections are **not**
  automatically populated as many of the others are:

  * `WEIGHTS`
  * `BIASES`
  * `ACTIVATIONS`
  """