org.apache.ignite.ml.nn

Class MLPTrainer<P extends Serializable>

java.lang.Object

org.apache.ignite.ml.trainers.DatasetTrainer<M,double[]>

org.apache.ignite.ml.trainers.MultiLabelDatasetTrainer<MultilayerPerceptron>

org.apache.ignite.ml.nn.MLPTrainer<P>

Type Parameters:

P - Type of model update used in this trainer.

public class MLPTrainer<P extends Serializable>
extends MultiLabelDatasetTrainer<MultilayerPerceptron>

Multilayer perceptron trainer based on partition based Dataset.

Nested Class Summary

Nested classes/interfaces inherited from class org.apache.ignite.ml.trainers.DatasetTrainer

DatasetTrainer.EmptyDatasetException

Field Summary

Fields inherited from class org.apache.ignite.ml.trainers.DatasetTrainer

envBuilder, environment

Constructor Summary

Constructors

Constructor and Description
MLPTrainer(IgniteFunction<Dataset<EmptyContext,SimpleLabeledDatasetData>,MLPArchitecture> archSupplier, IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction> loss, UpdatesStrategy<? super MultilayerPerceptron,P> updatesStgy, int maxIterations, int batchSize, int locIterations, long seed) Constructs a new instance of multilayer perceptron trainer.
MLPTrainer(MLPArchitecture arch, IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction> loss, UpdatesStrategy<? super MultilayerPerceptron,P> updatesStgy, int maxIterations, int batchSize, int locIterations, long seed) Constructs a new instance of multilayer perceptron trainer.

Method Summary

Modifier and Type	Method and Description
<K,V> MultilayerPerceptron	fitWithInitializedDeployingContext(DatasetBuilder<K,V> datasetBuilder, Preprocessor<K,V> extractor) Trains model based on the specified data.
IgniteFunction<Dataset<EmptyContext,SimpleLabeledDatasetData>,MLPArchitecture>	getArchSupplier() Get the multilayer perceptron architecture supplier that defines layers and activators.
int	getBatchSize() Get the batch size (per every partition).
int	getLocIterations() Get the maximal number of local iterations before synchronization.
IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction>	getLoss() Get the loss function to be minimized during the training.
int	getMaxIterations() Get the maximal number of iterations before the training will be stopped.
long	getSeed() Get the multilayer perceptron model initializer.
UpdatesStrategy<? super MultilayerPerceptron,P>	getUpdatesStgy() Get the update strategy that defines how to update model parameters during the training.
boolean	isUpdateable(MultilayerPerceptron mdl)
protected <K,V> MultilayerPerceptron	updateModel(MultilayerPerceptron lastLearnedMdl, DatasetBuilder<K,V> datasetBuilder, Preprocessor<K,V> extractor) Trains new model taken previous one as a first approximation.
MLPTrainer<P>	withArchSupplier(IgniteFunction<Dataset<EmptyContext,SimpleLabeledDatasetData>,MLPArchitecture> archSupplier) Set up the multilayer perceptron architecture supplier that defines layers and activators.
MLPTrainer<P>	withBatchSize(int batchSize) Set up the batch size (per every partition).
MLPTrainer<P>	withEnvironmentBuilder(LearningEnvironmentBuilder envBuilder) Changes learning Environment.
MLPTrainer<P>	withLocIterations(int locIterations) Set up the maximal number of local iterations before synchronization.
MLPTrainer<P>	withLoss(IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction> loss) Set up the loss function to be minimized during the training.
MLPTrainer<P>	withMaxIterations(int maxIterations) Set up the maximal number of iterations before the training will be stopped.
MLPTrainer<P>	withSeed(long seed) Set up the multilayer perceptron model initializer.
MLPTrainer<P>	withUpdatesStgy(UpdatesStrategy<? super MultilayerPerceptron,P> updatesStgy) Set up the update strategy that defines how to update model parameters during the training.

Methods inherited from class org.apache.ignite.ml.trainers.DatasetTrainer

fit, fit, fit, fit, fit, fit, getLastTrainedModelOrThrowEmptyDatasetException, identityTrainer, learningEnvironment, update, update, update, update, update, withConvertedLabels

Methods inherited from class java.lang.Object

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

Constructor Detail

MLPTrainer

public MLPTrainer(MLPArchitecture arch,
                  IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction> loss,
                  UpdatesStrategy<? super MultilayerPerceptron,P> updatesStgy,
                  int maxIterations,
                  int batchSize,
                  int locIterations,
                  long seed)

Constructs a new instance of multilayer perceptron trainer.

Parameters:

arch - Multilayer perceptron architecture that defines layers and activators.

loss - Loss function to be minimized during the training.

updatesStgy - Update strategy that defines how to update model parameters during the training.

maxIterations - Maximal number of iterations before the training will be stopped.

batchSize - Batch size (per every partition).

locIterations - Maximal number of local iterations before synchronization.

seed - Random initializer seed.

MLPTrainer

public MLPTrainer(IgniteFunction<Dataset<EmptyContext,SimpleLabeledDatasetData>,MLPArchitecture> archSupplier,
                  IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction> loss,
                  UpdatesStrategy<? super MultilayerPerceptron,P> updatesStgy,
                  int maxIterations,
                  int batchSize,
                  int locIterations,
                  long seed)

Constructs a new instance of multilayer perceptron trainer.

Parameters:

archSupplier - Multilayer perceptron architecture supplier that defines layers and activators.

loss - Loss function to be minimized during the training.

updatesStgy - Update strategy that defines how to update model parameters during the training.

maxIterations - Maximal number of iterations before the training will be stopped.

batchSize - Batch size (per every partition).

locIterations - Maximal number of local iterations before synchronization.

seed - Random initializer seed.

Method Detail

fitWithInitializedDeployingContext

public <K,V> MultilayerPerceptron fitWithInitializedDeployingContext(DatasetBuilder<K,V> datasetBuilder,
                                                                     Preprocessor<K,V> extractor)

Trains model based on the specified data.

Specified by:

fitWithInitializedDeployingContext in class DatasetTrainer<MultilayerPerceptron,double[]>

Type Parameters:

K - Type of a key in upstream data.

V - Type of a value in upstream data.

Parameters:

datasetBuilder - Dataset builder.

extractor - Extractor of UpstreamEntry into LabeledVector.

Returns:

Model.

updateModel

protected <K,V> MultilayerPerceptron updateModel(MultilayerPerceptron lastLearnedMdl,
                                                 DatasetBuilder<K,V> datasetBuilder,
                                                 Preprocessor<K,V> extractor)

Trains new model taken previous one as a first approximation.

Specified by:

updateModel in class DatasetTrainer<MultilayerPerceptron,double[]>

Type Parameters:

K - Type of a key in upstream data.

V - Type of a value in upstream data.

Parameters:

lastLearnedMdl - Learned model.

datasetBuilder - Dataset builder.

extractor - Extractor of UpstreamEntry into LabeledVector.

Returns:

Updated model.

getArchSupplier

public IgniteFunction<Dataset<EmptyContext,SimpleLabeledDatasetData>,MLPArchitecture> getArchSupplier()

Get the multilayer perceptron architecture supplier that defines layers and activators.

Returns:

The property value.

withArchSupplier

public MLPTrainer<P> withArchSupplier(IgniteFunction<Dataset<EmptyContext,SimpleLabeledDatasetData>,MLPArchitecture> archSupplier)

Set up the multilayer perceptron architecture supplier that defines layers and activators.

Parameters:

archSupplier - The parameter value.

Returns:

Model with the multilayer perceptron architecture supplier that defines layers and activators.

getLoss

public IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction> getLoss()

Get the loss function to be minimized during the training.

Returns:

The property value.

withLoss

public MLPTrainer<P> withLoss(IgniteFunction<Vector,IgniteDifferentiableVectorToDoubleFunction> loss)

Set up the loss function to be minimized during the training.

Parameters:

loss - The parameter value.

Returns:

Model with the loss function to be minimized during the training.

getUpdatesStgy

public UpdatesStrategy<? super MultilayerPerceptron,P> getUpdatesStgy()

Get the update strategy that defines how to update model parameters during the training.

Returns:

The property value.

withUpdatesStgy

public MLPTrainer<P> withUpdatesStgy(UpdatesStrategy<? super MultilayerPerceptron,P> updatesStgy)

Set up the update strategy that defines how to update model parameters during the training.

Parameters:

updatesStgy - The parameter value.

Returns:

Model with the update strategy that defines how to update model parameters during the training.

getMaxIterations

public int getMaxIterations()

Get the maximal number of iterations before the training will be stopped.

Returns:

The property value.

withMaxIterations

public MLPTrainer<P> withMaxIterations(int maxIterations)

Set up the maximal number of iterations before the training will be stopped.

Parameters:

maxIterations - The parameter value.

Returns:

Model with the maximal number of iterations before the training will be stopped.

getBatchSize

public int getBatchSize()

Get the batch size (per every partition).

Returns:

The property value.

withBatchSize

public MLPTrainer<P> withBatchSize(int batchSize)

Set up the batch size (per every partition).

Parameters:

batchSize - The parameter value.

Returns:

Model with the batch size (per every partition).

getLocIterations

public int getLocIterations()

Get the maximal number of local iterations before synchronization.

Returns:

The property value.

withLocIterations

public MLPTrainer<P> withLocIterations(int locIterations)

Set up the maximal number of local iterations before synchronization.

Parameters:

locIterations - The parameter value.

Returns:

Model with the maximal number of local iterations before synchronization.

getSeed

public long getSeed()

Get the multilayer perceptron model initializer.

Returns:

The property value.

withSeed

public MLPTrainer<P> withSeed(long seed)

Set up the multilayer perceptron model initializer.

Parameters:

seed - The parameter value.

Returns:

Model with the multilayer perceptron model initializer.

isUpdateable

public boolean isUpdateable(MultilayerPerceptron mdl)

Specified by:

isUpdateable in class DatasetTrainer<MultilayerPerceptron,double[]>

Parameters:

mdl - Model.

Returns:

True if current critical for training parameters correspond to parameters from last training.

withEnvironmentBuilder

public MLPTrainer<P> withEnvironmentBuilder(LearningEnvironmentBuilder envBuilder)

Changes learning Environment.

Overrides:

withEnvironmentBuilder in class DatasetTrainer<MultilayerPerceptron,double[]>

Parameters:

envBuilder - Learning environment builder.