org.deeplearning4j.spark.impl.paramavg

Class ParameterAveragingTrainingWorker

java.lang.Object

org.deeplearning4j.spark.impl.paramavg.ParameterAveragingTrainingWorker

All Implemented Interfaces:

java.io.Serializable, TrainingWorker<ParameterAveragingTrainingResult>

public class ParameterAveragingTrainingWorker
extends java.lang.Object
implements TrainingWorker<ParameterAveragingTrainingResult>

ParameterAveragingTrainingWorker implements standard parameter averaging every m iterations.

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
ParameterAveragingTrainingWorker(org.apache.spark.broadcast.Broadcast<NetBroadcastTuple> broadcast, boolean saveUpdater, WorkerConfiguration configuration, java.util.Collection<TrainingHook> trainingHooks, java.util.Collection<IterationListener> listeners, StatsStorageRouterProvider routerProvider)

Method Summary

Modifier and Type	Method and Description
void	addHook(TrainingHook trainingHook) Add a training hook to be used during training of the worker
WorkerConfiguration	getDataConfiguration() Get the WorkerConfiguration that contains information such as minibatch sizes, etc
ParameterAveragingTrainingResult	getFinalResult(ComputationGraph network) Get the final result to be returned to the driver
ParameterAveragingTrainingResult	getFinalResult(MultiLayerNetwork network) Get the final result to be returned to the driver
ParameterAveragingTrainingResult	getFinalResultNoData() Get the final result to be returned to the driver, if no data was available for this executor
Pair<ParameterAveragingTrainingResult,SparkTrainingStats>	getFinalResultNoDataWithStats() As per TrainingWorker.getFinalResultNoData() but used when SparkTrainingStats are being collected
Pair<ParameterAveragingTrainingResult,SparkTrainingStats>	getFinalResultWithStats(ComputationGraph graph) As per TrainingWorker.getFinalResult(ComputationGraph) but used when SparkTrainingStats are being collected
Pair<ParameterAveragingTrainingResult,SparkTrainingStats>	getFinalResultWithStats(MultiLayerNetwork network) As per TrainingWorker.getFinalResult(MultiLayerNetwork) but used when SparkTrainingStats are being collected
MultiLayerNetwork	getInitialModel() Get the initial model when training a MultiLayerNetwork/SparkDl4jMultiLayer
ComputationGraph	getInitialModelGraph() Get the initial model when training a ComputationGraph/SparkComputationGraph
ParameterAveragingTrainingResult	processMinibatch(org.nd4j.linalg.dataset.api.DataSet dataSet, ComputationGraph graph, boolean isLast) Process (fit) a minibatch for a ComputationGraph
ParameterAveragingTrainingResult	processMinibatch(org.nd4j.linalg.dataset.api.DataSet dataSet, MultiLayerNetwork network, boolean isLast) Process (fit) a minibatch for a MultiLayerNetwork
ParameterAveragingTrainingResult	processMinibatch(org.nd4j.linalg.dataset.api.MultiDataSet dataSet, ComputationGraph graph, boolean isLast) Process (fit) a minibatch for a ComputationGraph using a MultiDataSet
Pair<ParameterAveragingTrainingResult,SparkTrainingStats>	processMinibatchWithStats(org.nd4j.linalg.dataset.api.DataSet dataSet, ComputationGraph graph, boolean isLast) As per TrainingWorker.processMinibatch(DataSet, ComputationGraph, boolean) but used when SparkTrainingStats are being collected
Pair<ParameterAveragingTrainingResult,SparkTrainingStats>	processMinibatchWithStats(org.nd4j.linalg.dataset.api.DataSet dataSet, MultiLayerNetwork network, boolean isLast) As per TrainingWorker.processMinibatch(DataSet, MultiLayerNetwork, boolean) but used when SparkTrainingStats are being collecte
Pair<ParameterAveragingTrainingResult,SparkTrainingStats>	processMinibatchWithStats(org.nd4j.linalg.dataset.api.MultiDataSet dataSet, ComputationGraph graph, boolean isLast) As per TrainingWorker.processMinibatch(MultiDataSet, ComputationGraph, boolean) but used when SparkTrainingStats are being collected
void	removeHook(TrainingHook trainingHook) Remove a training hook from the worker

Methods inherited from class java.lang.Object

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

Constructor Detail

ParameterAveragingTrainingWorker

public ParameterAveragingTrainingWorker(org.apache.spark.broadcast.Broadcast<NetBroadcastTuple> broadcast,
                                        boolean saveUpdater,
                                        WorkerConfiguration configuration,
                                        java.util.Collection<TrainingHook> trainingHooks,
                                        java.util.Collection<IterationListener> listeners,
                                        StatsStorageRouterProvider routerProvider)

Method Detail

removeHook

public void removeHook(TrainingHook trainingHook)

Remove a training hook from the worker

Specified by:

removeHook in interface TrainingWorker<ParameterAveragingTrainingResult>

Parameters:

trainingHook - the training hook to remove

addHook

public void addHook(TrainingHook trainingHook)

Add a training hook to be used during training of the worker

Specified by:

addHook in interface TrainingWorker<ParameterAveragingTrainingResult>

Parameters:

trainingHook - the training hook to add

getInitialModel

public MultiLayerNetwork getInitialModel()

Description copied from interface: TrainingWorker

Get the initial model when training a MultiLayerNetwork/SparkDl4jMultiLayer

Specified by:

getInitialModel in interface TrainingWorker<ParameterAveragingTrainingResult>

Returns:

Initial model for this worker

getInitialModelGraph

public ComputationGraph getInitialModelGraph()

Description copied from interface: TrainingWorker

Get the initial model when training a ComputationGraph/SparkComputationGraph

Specified by:

getInitialModelGraph in interface TrainingWorker<ParameterAveragingTrainingResult>

Returns:

Initial model for this worker

processMinibatch

public ParameterAveragingTrainingResult processMinibatch(org.nd4j.linalg.dataset.api.DataSet dataSet,
                                                         MultiLayerNetwork network,
                                                         boolean isLast)

Description copied from interface: TrainingWorker

Process (fit) a minibatch for a MultiLayerNetwork

Specified by:

processMinibatch in interface TrainingWorker<ParameterAveragingTrainingResult>

Parameters:

dataSet - Data set to train on

network - Network to train

isLast - If true: last data set currently available. If false: more data sets will be processed for this executor

Returns:

Null, or a training result if training should be terminated immediately.

processMinibatch

public ParameterAveragingTrainingResult processMinibatch(org.nd4j.linalg.dataset.api.DataSet dataSet,
                                                         ComputationGraph graph,
                                                         boolean isLast)

Description copied from interface: TrainingWorker

Process (fit) a minibatch for a ComputationGraph

Specified by:

processMinibatch in interface TrainingWorker<ParameterAveragingTrainingResult>

Parameters:

dataSet - Data set to train on

graph - Network to train

isLast - If true: last data set currently available. If false: more data sets will be processed for this executor

Returns:

Null, or a training result if training should be terminated immediately.

processMinibatch

public ParameterAveragingTrainingResult processMinibatch(org.nd4j.linalg.dataset.api.MultiDataSet dataSet,
                                                         ComputationGraph graph,
                                                         boolean isLast)

Description copied from interface: TrainingWorker

Process (fit) a minibatch for a ComputationGraph using a MultiDataSet

Specified by:

processMinibatch in interface TrainingWorker<ParameterAveragingTrainingResult>

Parameters:

dataSet - Data set to train on

graph - Network to train

isLast - If true: last data set currently available. If false: more data sets will be processed for this executor

Returns:

Null, or a training result if training should be terminated immediately.

processMinibatchWithStats

public Pair<ParameterAveragingTrainingResult,SparkTrainingStats> processMinibatchWithStats(org.nd4j.linalg.dataset.api.DataSet dataSet,
                                                                                           MultiLayerNetwork network,
                                                                                           boolean isLast)

Description copied from interface: TrainingWorker

As per TrainingWorker.processMinibatch(DataSet, MultiLayerNetwork, boolean) but used when SparkTrainingStats are being collecte

Specified by:

processMinibatchWithStats in interface TrainingWorker<ParameterAveragingTrainingResult>

processMinibatchWithStats

public Pair<ParameterAveragingTrainingResult,SparkTrainingStats> processMinibatchWithStats(org.nd4j.linalg.dataset.api.DataSet dataSet,
                                                                                           ComputationGraph graph,
                                                                                           boolean isLast)

Description copied from interface: TrainingWorker

As per TrainingWorker.processMinibatch(DataSet, ComputationGraph, boolean) but used when SparkTrainingStats are being collected

Specified by:

processMinibatchWithStats in interface TrainingWorker<ParameterAveragingTrainingResult>

processMinibatchWithStats

public Pair<ParameterAveragingTrainingResult,SparkTrainingStats> processMinibatchWithStats(org.nd4j.linalg.dataset.api.MultiDataSet dataSet,
                                                                                           ComputationGraph graph,
                                                                                           boolean isLast)

Description copied from interface: TrainingWorker

As per TrainingWorker.processMinibatch(MultiDataSet, ComputationGraph, boolean) but used when SparkTrainingStats are being collected

Specified by:

processMinibatchWithStats in interface TrainingWorker<ParameterAveragingTrainingResult>

getFinalResult

public ParameterAveragingTrainingResult getFinalResult(MultiLayerNetwork network)

Description copied from interface: TrainingWorker

Get the final result to be returned to the driver

Specified by:

getFinalResult in interface TrainingWorker<ParameterAveragingTrainingResult>

Parameters:

network - Current state of the network

Returns:

Result to return to the driver

getFinalResult

public ParameterAveragingTrainingResult getFinalResult(ComputationGraph network)

Description copied from interface: TrainingWorker

Get the final result to be returned to the driver

Specified by:

getFinalResult in interface TrainingWorker<ParameterAveragingTrainingResult>

Parameters:

network - Current state of the network

Returns:

Result to return to the driver

getFinalResultNoData

public ParameterAveragingTrainingResult getFinalResultNoData()

Description copied from interface: TrainingWorker

Get the final result to be returned to the driver, if no data was available for this executor

Specified by:

getFinalResultNoData in interface TrainingWorker<ParameterAveragingTrainingResult>

Returns:

Result to return to the driver

getFinalResultNoDataWithStats

public Pair<ParameterAveragingTrainingResult,SparkTrainingStats> getFinalResultNoDataWithStats()

Description copied from interface: TrainingWorker

As per TrainingWorker.getFinalResultNoData() but used when SparkTrainingStats are being collected

Specified by:

getFinalResultNoDataWithStats in interface TrainingWorker<ParameterAveragingTrainingResult>

getFinalResultWithStats

public Pair<ParameterAveragingTrainingResult,SparkTrainingStats> getFinalResultWithStats(MultiLayerNetwork network)

Description copied from interface: TrainingWorker

As per TrainingWorker.getFinalResult(MultiLayerNetwork) but used when SparkTrainingStats are being collected

Specified by:

getFinalResultWithStats in interface TrainingWorker<ParameterAveragingTrainingResult>

getFinalResultWithStats

public Pair<ParameterAveragingTrainingResult,SparkTrainingStats> getFinalResultWithStats(ComputationGraph graph)

Description copied from interface: TrainingWorker

As per TrainingWorker.getFinalResult(ComputationGraph) but used when SparkTrainingStats are being collected

Specified by:

getFinalResultWithStats in interface TrainingWorker<ParameterAveragingTrainingResult>

getDataConfiguration

public WorkerConfiguration getDataConfiguration()

Description copied from interface: TrainingWorker

Get the WorkerConfiguration that contains information such as minibatch sizes, etc

Specified by:

getDataConfiguration in interface TrainingWorker<ParameterAveragingTrainingResult>

Returns:

Worker configuration