org.apache.ignite.ml.composition.boosting.convergence

Class ConvergenceChecker<K,V>

java.lang.Object

org.apache.ignite.ml.composition.boosting.convergence.ConvergenceChecker<K,V>

Type Parameters:

K - Type of a key in upstream data.

V - Type of a value in upstream data.

All Implemented Interfaces:

Serializable

Direct Known Subclasses:

ConvergenceCheckerStub, MeanAbsValueConvergenceChecker, MedianOfMedianConvergenceChecker

public abstract class ConvergenceChecker<K,V>
extends Object
implements Serializable

Contains logic of error computing and convergence checking for Gradient Boosting algorithms.

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
ConvergenceChecker(long sampleSize, IgniteFunction<Double,Double> externalLbToInternalMapping, Loss loss, DatasetBuilder<K,V> datasetBuilder, Preprocessor<K,V> preprocessor, double precision) Constructs an instance of ConvergenceChecker.

Method Summary

Modifier and Type	Method and Description
double	computeError(Vector features, Double answer, ModelsComposition currMdl) Compute error for the specific vector of dataset.
abstract Double	computeMeanErrorOnDataset(Dataset<EmptyContext,? extends FeatureMatrixWithLabelsOnHeapData> dataset, ModelsComposition mdl) Compute error for given model on learning dataset.
boolean	isConverged(Dataset<EmptyContext,? extends FeatureMatrixWithLabelsOnHeapData> dataset, ModelsComposition currMdl) Checks convergency on dataset.
boolean	isConverged(LearningEnvironmentBuilder envBuilder, DatasetBuilder<K,V> datasetBuilder, ModelsComposition currMdl) Checks convergency on dataset.

Methods inherited from class java.lang.Object

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

Constructor Detail

ConvergenceChecker

public ConvergenceChecker(long sampleSize,
                          IgniteFunction<Double,Double> externalLbToInternalMapping,
                          Loss loss,
                          DatasetBuilder<K,V> datasetBuilder,
                          Preprocessor<K,V> preprocessor,
                          double precision)

Constructs an instance of ConvergenceChecker.

Parameters:

sampleSize - Sample size.

externalLbToInternalMapping - External label to internal mapping.

loss - Loss gradient.

datasetBuilder - Dataset builder.

preprocessor - Upstream preprocessor.

precision - Precision.FeatureMatrixWithLabelsOnHeapDataBuilder.java

Method Detail

isConverged

public boolean isConverged(LearningEnvironmentBuilder envBuilder,
                           DatasetBuilder<K,V> datasetBuilder,
                           ModelsComposition currMdl)

Checks convergency on dataset.

Parameters:

envBuilder - Learning environment builder.

currMdl - Current model.

Returns:

True if GDB is converged.

isConverged

public boolean isConverged(Dataset<EmptyContext,? extends FeatureMatrixWithLabelsOnHeapData> dataset,
                           ModelsComposition currMdl)

Checks convergency on dataset.

Parameters:

dataset - Dataset.

currMdl - Current model.

Returns:

True if GDB is converged.

computeMeanErrorOnDataset

public abstract Double computeMeanErrorOnDataset(Dataset<EmptyContext,? extends FeatureMatrixWithLabelsOnHeapData> dataset,
                                                 ModelsComposition mdl)

Compute error for given model on learning dataset.

Parameters:

dataset - Learning dataset.

mdl - Model.

Returns:

Error mean value.

computeError

public double computeError(Vector features,
                           Double answer,
                           ModelsComposition currMdl)

Compute error for the specific vector of dataset.

Parameters:

currMdl - Current model.

Returns:

Error.