org.apache.ignite.ml.composition.boosting

Class GDBLearningStrategy

java.lang.Object

org.apache.ignite.ml.composition.boosting.GDBLearningStrategy

Direct Known Subclasses:

GDBOnTreesLearningStrategy

public class GDBLearningStrategy
extends Object

Learning strategy for gradient boosting.

Field Summary

Fields

Modifier and Type	Field and Description
protected IgniteSupplier<DatasetTrainer<? extends Model<Vector,Double>,Double>>	baseMdlTrainerBuilder Base model trainer builder.
protected ConvergenceCheckerFactory	checkConvergenceStgyFactory Check convergence strategy factory.
protected int	cntOfIterations Count of iterations.
protected double[]	compositionWeights Composition weights.
protected LearningEnvironment	environment Learning environment.
protected IgniteFunction<Double,Double>	externalLbToInternalMapping External label to internal mapping.
protected Loss	loss Loss of gradient.
protected double	meanLbVal Mean label value.
protected long	sampleSize Sample size.

Constructor Summary

Constructors

Constructor and Description
GDBLearningStrategy()

Method Summary

Modifier and Type	Method and Description
double[]	getCompositionWeights()
double	getMeanValue()
protected @NotNull List<Model<Vector,Double>>	initLearningState(GDBTrainer.GDBModel mdlToUpdate) Restores state of already learned model if can and sets learning parameters according to this state.
<K,V> List<Model<Vector,Double>>	learnModels(DatasetBuilder<K,V> datasetBuilder, IgniteBiFunction<K,V,Vector> featureExtractor, IgniteBiFunction<K,V,Double> lbExtractor) Implementation of gradient boosting iterations.
<K,V> List<Model<Vector,Double>>	update(GDBTrainer.GDBModel mdlToUpdate, DatasetBuilder<K,V> datasetBuilder, IgniteBiFunction<K,V,Vector> featureExtractor, IgniteBiFunction<K,V,Double> lbExtractor) Gets state of model in arguments, compare it with training parameters of trainer and if they are fit then trainer updates model in according to new data and return new model.
GDBLearningStrategy	withBaseModelTrainerBuilder(IgniteSupplier<DatasetTrainer<? extends Model<Vector,Double>,Double>> buildBaseMdlTrainer) Sets base model builder.
GDBLearningStrategy	withCheckConvergenceStgyFactory(ConvergenceCheckerFactory factory) Sets CheckConvergenceStgyFactory.
GDBLearningStrategy	withCntOfIterations(int cntOfIterations) Sets count of iterations.
GDBLearningStrategy	withCompositionWeights(double[] compositionWeights) Sets composition weights vector.
GDBLearningStrategy	withDefaultGradStepSize(double defaultGradStepSize) Sets default gradient step size.
GDBLearningStrategy	withEnvironment(LearningEnvironment environment) Sets learning environment.
GDBLearningStrategy	withExternalLabelToInternal(IgniteFunction<Double,Double> externalLbToInternal) Sets external to internal label representation mapping.
GDBLearningStrategy	withLossGradient(Loss loss) Loss function.
GDBLearningStrategy	withMeanLabelValue(double meanLbVal) Sets mean label value.
GDBLearningStrategy	withSampleSize(long sampleSize) Sets sample size.

Methods inherited from class java.lang.Object

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

Field Detail

environment

protected LearningEnvironment environment

Learning environment.

cntOfIterations

protected int cntOfIterations

Count of iterations.

loss

protected Loss loss

Loss of gradient.

externalLbToInternalMapping

protected IgniteFunction<Double,Double> externalLbToInternalMapping

External label to internal mapping.

baseMdlTrainerBuilder

protected IgniteSupplier<DatasetTrainer<? extends Model<Vector,Double>,Double>> baseMdlTrainerBuilder

Base model trainer builder.

meanLbVal

protected double meanLbVal

Mean label value.

sampleSize

protected long sampleSize

Sample size.

compositionWeights

protected double[] compositionWeights

Composition weights.

checkConvergenceStgyFactory

protected ConvergenceCheckerFactory checkConvergenceStgyFactory

Check convergence strategy factory.

Constructor Detail

GDBLearningStrategy

public GDBLearningStrategy()

Method Detail

learnModels

public <K,V> List<Model<Vector,Double>> learnModels(DatasetBuilder<K,V> datasetBuilder,
                                                    IgniteBiFunction<K,V,Vector> featureExtractor,
                                                    IgniteBiFunction<K,V,Double> lbExtractor)

Implementation of gradient boosting iterations. At each step of iterations this algorithm build a regression model based on gradient of loss-function for current models composition.

Parameters:

datasetBuilder - Dataset builder.

featureExtractor - Feature extractor.

lbExtractor - Label extractor.

Returns:

list of learned models.

update

public <K,V> List<Model<Vector,Double>> update(GDBTrainer.GDBModel mdlToUpdate,
                                               DatasetBuilder<K,V> datasetBuilder,
                                               IgniteBiFunction<K,V,Vector> featureExtractor,
                                               IgniteBiFunction<K,V,Double> lbExtractor)

Gets state of model in arguments, compare it with training parameters of trainer and if they are fit then trainer updates model in according to new data and return new model. In other case trains new model.

Type Parameters:

K - Type of a key in upstream data.

V - Type of a value in upstream data.

Parameters:

mdlToUpdate - Learned model.

datasetBuilder - Dataset builder.

featureExtractor - Feature extractor.

lbExtractor - Label extractor.

Returns:

Updated models list.

initLearningState

@NotNull
protected @NotNull List<Model<Vector,Double>> initLearningState(GDBTrainer.GDBModel mdlToUpdate)

Restores state of already learned model if can and sets learning parameters according to this state.

Parameters:

mdlToUpdate - Model to update.

Returns:

list of already learned models.

withEnvironment

public GDBLearningStrategy withEnvironment(LearningEnvironment environment)

Sets learning environment.

Parameters:

environment - Learning Environment.

withCntOfIterations

public GDBLearningStrategy withCntOfIterations(int cntOfIterations)

Sets count of iterations.

Parameters:

cntOfIterations - Count of iterations.

withLossGradient

public GDBLearningStrategy withLossGradient(Loss loss)

Loss function.

Parameters:

loss - Loss function.

withExternalLabelToInternal

public GDBLearningStrategy withExternalLabelToInternal(IgniteFunction<Double,Double> externalLbToInternal)

Sets external to internal label representation mapping.

Parameters:

externalLbToInternal - External label to internal.

withBaseModelTrainerBuilder

public GDBLearningStrategy withBaseModelTrainerBuilder(IgniteSupplier<DatasetTrainer<? extends Model<Vector,Double>,Double>> buildBaseMdlTrainer)

Sets base model builder.

Parameters:

buildBaseMdlTrainer - Build base model trainer.

withMeanLabelValue

public GDBLearningStrategy withMeanLabelValue(double meanLbVal)

Sets mean label value.

Parameters:

meanLbVal - Mean label value.

withSampleSize

public GDBLearningStrategy withSampleSize(long sampleSize)

Sets sample size.

Parameters:

sampleSize - Sample size.

withCompositionWeights

public GDBLearningStrategy withCompositionWeights(double[] compositionWeights)

Sets composition weights vector.

Parameters:

compositionWeights - Composition weights.

withCheckConvergenceStgyFactory

public GDBLearningStrategy withCheckConvergenceStgyFactory(ConvergenceCheckerFactory factory)

Sets CheckConvergenceStgyFactory.

Parameters:

factory - Factory.

withDefaultGradStepSize

public GDBLearningStrategy withDefaultGradStepSize(double defaultGradStepSize)

Sets default gradient step size.

Parameters:

defaultGradStepSize - Default gradient step size.

getCompositionWeights

public double[] getCompositionWeights()

getMeanValue

public double getMeanValue()