com.feedzai.openml.provider.lightgbm

Class SWIGTrainData

java.lang.Object

com.feedzai.openml.provider.lightgbm.SWIGTrainData

All Implemented Interfaces:

AutoCloseable

public class SWIGTrainData
extends Object
implements AutoCloseable

Handles train data resources and provides basic operations to manipulate train data. This class is responsible for initializing, managing and releasing all Handles train data resources and provides basic operations to manipulate train data. LightGBM SWIG train resources and resource handlers in a memory-safe manner. Whatever happens, it guarantees that no memory leaks are left behind.

Author:

Alberto Ferreira (alberto.ferreira@feedzai.com)

Since:

1.0.10

Field Summary

Fields

Modifier and Type	Field and Description
boolean	fairnessConstrained Whether the LightGBM model is fairness constrained (aka FairGBM).
int	numFeatures Number of features per instance.

Constructor Summary

Constructors

Constructor and Description
SWIGTrainData(int numFeatures, long numInstancesChunk) Constructor.
SWIGTrainData(int numFeatures, long numInstancesChunk, boolean fairnessConstrained) Constructor.

Method Summary

Modifier and Type	Method and Description
void	addConstraintGroupValue(int value) Adds a value to the constraint group ChunkedArray.
void	addFeatureValue(double value) Adds a value to the features' ChunkedArray.
void	addLabelValue(float value) Adds a value to the labels' ChunkedArray.
void	close() Release any allocated resources.
long	getNumInstancesChunk()

Methods inherited from class java.lang.Object

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

Field Detail

numFeatures

public final int numFeatures

Number of features per instance.

fairnessConstrained

public final boolean fairnessConstrained

Whether the LightGBM model is fairness constrained (aka FairGBM).

Constructor Detail

SWIGTrainData

public SWIGTrainData(int numFeatures,
                     long numInstancesChunk)

Constructor. Allocates all the initial handles necessary to bootstrap (but not use) the in-memory LightGBM dataset + booster structures. After that the BoosterHandle and the DatasetHandle will still need to be initialized at the proper times:

Parameters:

numFeatures - The number of features.

numInstancesChunk - The number of instances per chunk of data.

See Also:

initSwigDatasetHandle()

SWIGTrainData

public SWIGTrainData(int numFeatures,
                     long numInstancesChunk,
                     boolean fairnessConstrained)

Constructor. Allocates al the initial ahndles necessary to bootstrap (but not use) the in-memory LightGBM dataset, plus booster structures. If fairnessConstrained=true, this will also include data on which sensitive group each instance belongs to.

Parameters:

numFeatures - The number of features.

numInstancesChunk - The number of instances per chunk of data.

fairnessConstrained - Whether this data will be used for a model with fairness (group) constraints.

Method Detail

addFeatureValue

public void addFeatureValue(double value)

Adds a value to the features' ChunkedArray.

Parameters:

value - value to insert.

addLabelValue

public void addLabelValue(float value)

Adds a value to the labels' ChunkedArray.

addConstraintGroupValue

public void addConstraintGroupValue(int value)

Adds a value to the constraint group ChunkedArray.

Parameters:

value - the value to add.

getNumInstancesChunk

public long getNumInstancesChunk()

Returns:

Return the chunk sizes (in instances).

close

public void close()

Release any allocated resources. This operation is idempotent and can be safely called at any time as many times as you wish.

Specified by:

close in interface AutoCloseable