Module: tf.train

Defined in tensorflow/train/

Support for training models.

See the Training guide.


queue_runner module: Create threads to run multiple enqueue ops.


class AdadeltaOptimizer: Optimizer that implements the Adadelta algorithm.

class AdagradDAOptimizer: Adagrad Dual Averaging algorithm for sparse linear models.

class AdagradOptimizer: Optimizer that implements the Adagrad algorithm.

class AdamOptimizer: Optimizer that implements the Adam algorithm.

class BytesList: A ProtocolMessage

class Checkpoint: Groups checkpointable objects, saving and restoring them.

class CheckpointSaverHook: Saves checkpoints every N steps or seconds.

class CheckpointSaverListener: Interface for listeners that take action before or after checkpoint save.

class ChiefSessionCreator: Creates a tf.Session for a chief.

class ClusterDef: A ProtocolMessage

class ClusterSpec: Represents a cluster as a set of "tasks", organized into "jobs".

class Coordinator: A coordinator for threads.

class Example: A ProtocolMessage

class ExponentialMovingAverage: Maintains moving averages of variables by employing an exponential decay.

class Feature: A ProtocolMessage

class FeatureList: A ProtocolMessage

class FeatureLists: A ProtocolMessage

class Features: A ProtocolMessage

class FeedFnHook: Runs feed_fn and sets the feed_dict accordingly.

class FinalOpsHook: A hook which evaluates Tensors at the end of a session.

class FloatList: A ProtocolMessage

class FtrlOptimizer: Optimizer that implements the FTRL algorithm.

class GlobalStepWaiterHook: Delays execution until global step reaches wait_until_step.

class GradientDescentOptimizer: Optimizer that implements the gradient descent algorithm.

class Int64List: A ProtocolMessage

class JobDef: A ProtocolMessage

class LoggingTensorHook: Prints the given tensors every N local steps, every N seconds, or at end.

class LooperThread: A thread that runs code repeatedly, optionally on a timer.

class MomentumOptimizer: Optimizer that implements the Momentum algorithm.

class MonitoredSession: Session-like object that handles initialization, recovery and hooks.

class NanLossDuringTrainingError: Unspecified run-time error.

class NanTensorHook: Monitors the loss tensor and stops training if loss is NaN.

class Optimizer: Base class for optimizers.

class ProfilerHook: Captures CPU/GPU profiling information every N steps or seconds.

class ProximalAdagradOptimizer: Optimizer that implements the Proximal Adagrad algorithm.

class ProximalGradientDescentOptimizer: Optimizer that implements the proximal gradient descent algorithm.

class QueueRunner: Holds a list of enqueue operations for a queue, each to be run in a thread.

class RMSPropOptimizer: Optimizer that implements the RMSProp algorithm.

class Saver: Saves and restores variables.

class SaverDef: A ProtocolMessage

class Scaffold: Structure to create or gather pieces commonly needed to train a model.

class SecondOrStepTimer: Timer that triggers at most once every N seconds or once every N steps.

class SequenceExample: A ProtocolMessage

class Server: An in-process TensorFlow server, for use in distributed training.

class ServerDef: A ProtocolMessage

class SessionCreator: A factory for tf.Session.

class SessionManager: Training helper that restores from checkpoint and creates session.

class SessionRunArgs: Represents arguments to be added to a call.

class SessionRunContext: Provides information about the call being made.

class SessionRunHook: Hook to extend calls to

class SessionRunValues: Contains the results of

class SingularMonitoredSession: Session-like object that handles initialization, restoring, and hooks.

class StepCounterHook: Hook that counts steps per second.

class StopAtStepHook: Hook that requests stop at a specified step.

class SummarySaverHook: Saves summaries every N steps.

class Supervisor: A training helper that checkpoints models and computes summaries.

class SyncReplicasOptimizer: Class to synchronize, aggregate gradients and pass them to the optimizer.

class VocabInfo: Vocabulary information for warm-starting.

class WorkerSessionCreator: Creates a tf.Session for a worker.


MonitoredTrainingSession(...): Creates a MonitoredSession for training.


add_queue_runner(...): Adds a QueueRunner to a collection in the graph.

assert_global_step(...): Asserts global_step_tensor is a scalar int Variable or Tensor.

basic_train_loop(...): Basic loop to train a model.

batch(...): Creates batches of tensors in tensors.

batch_join(...): Runs a list of tensors to fill a queue to create batches of examples.

checkpoint_exists(...): Checks whether a V1 or V2 checkpoint exists with the specified prefix.

cosine_decay(...): Applies cosine decay to the learning rate.

cosine_decay_restarts(...): Applies cosine decay with restarts to the learning rate.

create_global_step(...): Create global step tensor in graph.

do_quantize_training_on_graphdef(...): A general quantization scheme is being developed in tf.contrib.quantize.

exponential_decay(...): Applies exponential decay to the learning rate.

export_meta_graph(...): Returns MetaGraphDef proto. Optionally writes it to filename.

generate_checkpoint_state_proto(...): Generates a checkpoint state proto.

get_checkpoint_mtimes(...): Returns the mtimes (modification timestamps) of the checkpoints.

get_checkpoint_state(...): Returns CheckpointState proto from the "checkpoint" file.

get_global_step(...): Get the global step tensor.

get_or_create_global_step(...): Returns and create (if necessary) the global step tensor.

global_step(...): Small helper to get the global step.

import_meta_graph(...): Recreates a Graph saved in a MetaGraphDef proto.

init_from_checkpoint(...): Initializes current variables with tensors loaded from given checkpoint.

input_producer(...): Output the rows of input_tensor to a queue for an input pipeline.

inverse_time_decay(...): Applies inverse time decay to the initial learning rate.

latest_checkpoint(...): Finds the filename of latest saved checkpoint file.

limit_epochs(...): Returns tensor num_epochs times and then raises an OutOfRange error.

linear_cosine_decay(...): Applies linear cosine decay to the learning rate.

list_variables(...): Returns list of all variables in the checkpoint.

load_checkpoint(...): Returns CheckpointReader for checkpoint found in ckpt_dir_or_file.

load_variable(...): Returns the tensor value of the given variable in the checkpoint.

match_filenames_once(...): Save the list of files matching pattern, so it is only computed once.

maybe_batch(...): Conditionally creates batches of tensors based on keep_input.

maybe_batch_join(...): Runs a list of tensors to conditionally fill a queue to create batches.

maybe_shuffle_batch(...): Creates batches by randomly shuffling conditionally-enqueued tensors.

maybe_shuffle_batch_join(...): Create batches by randomly shuffling conditionally-enqueued tensors.

natural_exp_decay(...): Applies natural exponential decay to the initial learning rate.

noisy_linear_cosine_decay(...): Applies noisy linear cosine decay to the learning rate.

piecewise_constant(...): Piecewise constant from boundaries and interval values.

polynomial_decay(...): Applies a polynomial decay to the learning rate.

range_input_producer(...): Produces the integers from 0 to limit-1 in a queue.

replica_device_setter(...): Return a device function to use when building a Graph for replicas.

sdca_fprint(...): Computes fingerprints of the input strings.

sdca_optimizer(...): Distributed version of Stochastic Dual Coordinate Ascent (SDCA) optimizer for

sdca_shrink_l1(...): Applies L1 regularization shrink step on the parameters.

shuffle_batch(...): Creates batches by randomly shuffling tensors.

shuffle_batch_join(...): Create batches by randomly shuffling tensors.

slice_input_producer(...): Produces a slice of each Tensor in tensor_list.

start_queue_runners(...): Starts all queue runners collected in the graph.

string_input_producer(...): Output strings (e.g. filenames) to a queue for an input pipeline.

summary_iterator(...): An iterator for reading Event protocol buffers from an event file.

update_checkpoint_state(...): Updates the content of the 'checkpoint' file.

warm_start(...): Warm-starts a model using the given settings.

write_graph(...): Writes a graph proto to a file.

Other Members