View source on GitHub
|
Training and input utilities.
See Contrib Training guide.
Classes
class FeedingQueueRunner: A queue runner that allows the feeding of values such as numpy arrays.
class GreedyLoadBalancingStrategy: Returns the least-loaded ps task for op placement.
class HParams: Class to hold a set of hyperparameters as name-value pairs.
class NextQueuedSequenceBatch: NextQueuedSequenceBatch stores deferred SequenceQueueingStateSaver data.
class RandomStrategy: Returns a random PS task for op placement.
class SequenceQueueingStateSaver: SequenceQueueingStateSaver provides access to stateful values from input.
class StopAfterNEvalsHook: Run hook used by the evaluation routines to run the eval_ops N times.
class SummaryAtEndHook: A run hook that saves a summary with the results of evaluation.
Functions
add_gradients_summaries(...): Add summaries to gradients.
batch_sequences_with_states(...): Creates batches of segments of sequential input.
bucket(...): Lazy bucketing of input tensors according to which_bucket.
bucket_by_sequence_length(...): Lazy bucketing of inputs according to their length.
byte_size_load_fn(...): Load function that computes the byte size of a single-output Operation.
checkpoints_iterator(...): Continuously yield new checkpoint files as they appear.
clip_gradient_norms(...): Clips the gradients by the given value.
clip_gradient_norms_fn(...): Returns a transform_grads_fn function for gradient clipping.
create_train_op(...): Creates an Operation that evaluates the gradients and returns the loss.
evaluate_once(...): Evaluates the model at the given checkpoint path.
evaluate_repeatedly(...): Repeatedly searches for a checkpoint in checkpoint_dir and evaluates it.
get_or_create_eval_step(...): Gets or creates the eval step Tensor.
multiply_gradients(...): Multiply specified gradients.
parse_values(...): Parses hyperparameter values from a string into a python map.
rejection_sample(...): Stochastically creates batches by rejection sampling.
resample_at_rate(...): Given inputs tensors, stochastically resamples each at a given rate.
stratified_sample(...): Stochastically creates batches based on per-class probabilities.
train(...): Runs the training loop.
wait_for_new_checkpoint(...): Waits until a new checkpoint file is found.
weighted_resample(...): Performs an approximate weighted resampling of inputs.