org.tribuo.interop.tensorflow.sequence

Class TensorFlowSequenceTrainer<T extends Output<T>>

java.lang.Object

org.tribuo.interop.tensorflow.sequence.TensorFlowSequenceTrainer<T>

All Implemented Interfaces:

com.oracle.labs.mlrg.olcut.config.Configurable, com.oracle.labs.mlrg.olcut.provenance.Provenancable<TrainerProvenance>, SequenceTrainer<T>

public class TensorFlowSequenceTrainer<T extends Output<T>>
extends Object
implements SequenceTrainer<T>

A trainer for SequenceModels which use an underlying TensorFlow graph.

N.B. TensorFlow support is experimental and may change without a major version bump.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	TensorFlowSequenceTrainer.TensorFlowSequenceTrainerProvenance

Field Summary

Fields

Modifier and Type	Field and Description
protected int	epochs
protected SequenceFeatureConverter	featureConverter
protected String	getLossOp
protected Path	graphPath
protected int	loggingInterval
protected int	minibatchSize
protected SequenceOutputConverter<T>	outputConverter
protected String	predictOp
protected SplittableRandom	rng
protected long	seed
protected int	trainInvocationCounter
protected String	trainOp

Constructor Summary

Constructors

Constructor and Description
TensorFlowSequenceTrainer(Path graphPath, SequenceFeatureConverter featureConverter, SequenceOutputConverter<T> outputConverter, int minibatchSize, int epochs, int loggingInterval, long seed, String trainOp, String getLossOp, String predictOp) Constructs a TensorFlowSequenceTrainer using the specified parameters.

Method Summary

Modifier and Type	Method and Description
protected TensorMap	getHyperparameterFeed() Build any necessary non-data parameter tensors.
int	getInvocationCount() Returns the number of times the train method has been invoked.
TrainerProvenance	getProvenance()
void	postConfig() Used by the OLCUT configuration system, and should not be called by external code.
protected void	preTrainingHook(org.tensorflow.Session session, SequenceDataset<T> examples) A hook for modifying the session state before training starts.
String	toString()
SequenceModel<T>	train(SequenceDataset<T> examples, Map<String,com.oracle.labs.mlrg.olcut.provenance.Provenance> runProvenance) Trains a sequence prediction model using the examples in the given data set.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Methods inherited from interface org.tribuo.sequence.SequenceTrainer

train

Field Detail

graphPath

@Config(mandatory=true,
        description="Path to the protobuf containing the TensorFlow graph.")
protected Path graphPath

featureConverter

@Config(mandatory=true,
        description="Sequence feature extractor.")
protected SequenceFeatureConverter featureConverter

outputConverter

@Config(mandatory=true,
        description="Sequence output extractor.")
protected SequenceOutputConverter<T extends Output<T>> outputConverter

minibatchSize

@Config(description="Minibatch size")
protected int minibatchSize

epochs

@Config(description="Number of SGD epochs to run.")
protected int epochs

loggingInterval

@Config(description="Logging interval to print the loss.")
protected int loggingInterval

seed

@Config(description="Seed for the RNG.")
protected long seed

trainOp

@Config(mandatory=true,
        description="Name of the training operation.")
protected String trainOp

getLossOp

@Config(mandatory=true,
        description="Name of the loss operation (to inspect the loss).")
protected String getLossOp

predictOp

@Config(mandatory=true,
        description="Name of the prediction operation.")
protected String predictOp

rng

protected SplittableRandom rng

trainInvocationCounter

protected int trainInvocationCounter

Constructor Detail

TensorFlowSequenceTrainer

public TensorFlowSequenceTrainer(Path graphPath,
                                 SequenceFeatureConverter featureConverter,
                                 SequenceOutputConverter<T> outputConverter,
                                 int minibatchSize,
                                 int epochs,
                                 int loggingInterval,
                                 long seed,
                                 String trainOp,
                                 String getLossOp,
                                 String predictOp)
                          throws IOException

Constructs a TensorFlowSequenceTrainer using the specified parameters.

Parameters:

graphPath - The path to the TF graph.

featureConverter - The feature conversion object.

outputConverter - The output conversion object.

minibatchSize - The training minibatch size.

epochs - The number of training epochs.

loggingInterval - The logging interval.

seed - The RNG seed.

trainOp - The name of the training operation.

getLossOp - The name of the loss operation.

predictOp - The name of the prediction operation.

Throws:

IOException - If the graph can not be read from disk.

Method Detail

postConfig

public void postConfig()
                throws IOException

Used by the OLCUT configuration system, and should not be called by external code.

Specified by:

postConfig in interface com.oracle.labs.mlrg.olcut.config.Configurable

Throws:

IOException

train

public SequenceModel<T> train(SequenceDataset<T> examples,
                              Map<String,com.oracle.labs.mlrg.olcut.provenance.Provenance> runProvenance)

Description copied from interface: SequenceTrainer

Trains a sequence prediction model using the examples in the given data set.

Specified by:

train in interface SequenceTrainer<T extends Output<T>>

Parameters:

examples - the data set containing the examples.

runProvenance - Training run specific provenance (e.g., fold number).

Returns:

a predictive model that can be used to generate predictions for new examples.

getInvocationCount

public int getInvocationCount()

Description copied from interface: SequenceTrainer

Returns the number of times the train method has been invoked.

Specified by:

getInvocationCount in interface SequenceTrainer<T extends Output<T>>

Returns:

The number of times train has been invoked.

toString

public String toString()

Overrides:

toString in class Object

preTrainingHook

protected void preTrainingHook(org.tensorflow.Session session,
                               SequenceDataset<T> examples)

A hook for modifying the session state before training starts.

This should not mutate any state in the trainer.

Parameters:

session - The session to modify.

examples - The dataset.

getHyperparameterFeed

protected TensorMap getHyperparameterFeed()

Build any necessary non-data parameter tensors.

The default implementation returns an empty map, and should be overridden if necessary.

This should not mutate any state in the trainer.

Returns:

The parameter tensors.

getProvenance

public TrainerProvenance getProvenance()

Specified by:

getProvenance in interface com.oracle.labs.mlrg.olcut.provenance.Provenancable<TrainerProvenance>