org.apache.ignite.ml.recommendation.util

Class RecommendationDatasetData<O extends Serializable,S extends Serializable>

java.lang.Object

org.apache.ignite.ml.recommendation.util.RecommendationDatasetData<O,S>

Type Parameters:

O - Type of an object.

S - Type of a subject.

All Implemented Interfaces:

AutoCloseable

public class RecommendationDatasetData<O extends Serializable,S extends Serializable>
extends Object
implements AutoCloseable

A partition data of a dataset required in RecommendationTrainer.

Constructor Summary

Constructors

Constructor and Description
RecommendationDatasetData(List<? extends ObjectSubjectRatingTriplet<O,S>> ratings) Constructs a new instance of recommendation dataset data.

Method Summary

Modifier and Type	Method and Description
MatrixFactorizationGradient<O,S>	calculateGradient(Map<O,Vector> objMatrix, Map<S,Vector> subjMatrix, int batchSize, int seed, double regParam, double learningRate) Calculates gradient of the loss function of recommendation system SGD training.
void	close()
Set<O>	getObjects() Returns set of objects contained in ratings.
Set<S>	getSubjects() Returns set of subjects contained in ratings.

Methods inherited from class java.lang.Object

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

Constructor Detail

RecommendationDatasetData

public RecommendationDatasetData(List<? extends ObjectSubjectRatingTriplet<O,S>> ratings)

Constructs a new instance of recommendation dataset data.

Parameters:

ratings - All ratings related to the partition.

Method Detail

calculateGradient

public MatrixFactorizationGradient<O,S> calculateGradient(Map<O,Vector> objMatrix,
                                                          Map<S,Vector> subjMatrix,
                                                          int batchSize,
                                                          int seed,
                                                          double regParam,
                                                          double learningRate)

Calculates gradient of the loss function of recommendation system SGD training. The details about gradient calculation could be found here: https://tinyurl.com/y6cku9hr.

Parameters:

objMatrix - Object matrix obtained as a result of factorization of rating matrix.

subjMatrix - Subject matrix obtained as a result of factorization of rating matrix.

batchSize - Batch size of stochastic gradient descent. The size of a dataset used on each step of SGD.

seed - Seed (required to make randomized part behaviour repeatable).

regParam - Regularization parameter.

learningRate - Learning rate.

Returns:

Gradient of matrix factorization loss function.

getObjects

public Set<O> getObjects()

Returns set of objects contained in ratings.

Returns:

Set of object.

getSubjects

public Set<S> getSubjects()

Returns set of subjects contained in ratings.

Returns:

Set of subjects.

close

public void close()

Specified by:

close in interface AutoCloseable