View source on GitHub |
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 ofVariable
objects, shared across distributed environment (model variables are subset of these). Seetf.compat.v1.global_variables
for more details. Commonly, allTRAINABLE_VARIABLES
variables will be inMODEL_VARIABLES
, and allMODEL_VARIABLES
variables will be inGLOBAL_VARIABLES
.LOCAL_VARIABLES
: the subset ofVariable
objects that are local to each machine. Usually used for temporarily variables, like counters. Note: usetf.contrib.framework.local_variable
to add to this collection.MODEL_VARIABLES
: the subset ofVariable
objects that are used in the model for inference (feed forward). Note: usetf.contrib.framework.model_variable
to add to this collection.TRAINABLE_VARIABLES
: the subset ofVariable
objects that will be trained by an optimizer. Seetf.compat.v1.trainable_variables
for more details.SUMMARIES
: the summaryTensor
objects that have been created in the graph. Seetf.compat.v1.summary.merge_all
for more details.QUEUE_RUNNERS
: theQueueRunner
objects that are used to produce input for a computation. Seetf.compat.v1.train.start_queue_runners
for more details.MOVING_AVERAGE_VARIABLES
: the subset ofVariable
objects that will also keep moving averages. Seetf.compat.v1.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