es.uam.eps.ir.ranksys.mf.als

Class HKVFactorizer<U,I>

java.lang.Object

es.uam.eps.ir.ranksys.mf.Factorizer<U,I>

es.uam.eps.ir.ranksys.mf.als.ALSFactorizer<U,I>

es.uam.eps.ir.ranksys.mf.als.HKVFactorizer<U,I>

Type Parameters:

U - type of the users

I - type of the items

public class HKVFactorizer<U,I>
extends ALSFactorizer<U,I>

Implicit matrix factorization of Hu, Koren and Volinsky. Y. Hu, Y. Koren, C. Volinsky. Collaborative filtering for implicit feedback datasets. ICDM 2008.

Author:

Saúl Vargas (saul.vargas@uam.es)

Constructor Summary

Constructors

Constructor and Description
HKVFactorizer(double lambdaP, double lambdaQ, DoubleUnaryOperator confidence, int numIter) Constructor.
HKVFactorizer(double lambda, DoubleUnaryOperator confidence, int numIter) Constructor.

Method Summary

Modifier and Type	Method and Description
double	error(cern.colt.matrix.impl.DenseDoubleMatrix2D p, cern.colt.matrix.impl.DenseDoubleMatrix2D q, FastPreferenceData<U,I> data) Squared loss of two matrices.
void	set_minP(cern.colt.matrix.impl.DenseDoubleMatrix2D p, cern.colt.matrix.impl.DenseDoubleMatrix2D q, FastPreferenceData<U,I> data) User matrix least-squares step.
void	set_minQ(cern.colt.matrix.impl.DenseDoubleMatrix2D q, cern.colt.matrix.impl.DenseDoubleMatrix2D p, FastPreferenceData<U,I> data) Item matrix least-squares step.

Methods inherited from class es.uam.eps.ir.ranksys.mf.als.ALSFactorizer

error, factorize, factorize

Methods inherited from class java.lang.Object

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

Constructor Detail

HKVFactorizer

public HKVFactorizer(double lambda,
                     DoubleUnaryOperator confidence,
                     int numIter)

Constructor. Same regularization factor for user and item matrices.

Parameters:

lambda - regularization factor

confidence - confidence function

numIter - number of iterations

HKVFactorizer

public HKVFactorizer(double lambdaP,
                     double lambdaQ,
                     DoubleUnaryOperator confidence,
                     int numIter)

Constructor. Different regularization factors for user and item matrices.

Parameters:

lambdaP - regularization factor for user matrix

lambdaQ - regularization factor for item matrix

confidence - confidence function

numIter - number of iterations

Method Detail

error

public double error(cern.colt.matrix.impl.DenseDoubleMatrix2D p,
                    cern.colt.matrix.impl.DenseDoubleMatrix2D q,
                    FastPreferenceData<U,I> data)

Description copied from class: ALSFactorizer

Squared loss of two matrices.

Specified by:

error in class ALSFactorizer<U,I>

Parameters:

p - user matrix

q - item matrix

data - preference data

Returns:

squared loss

set_minP

public void set_minP(cern.colt.matrix.impl.DenseDoubleMatrix2D p,
                     cern.colt.matrix.impl.DenseDoubleMatrix2D q,
                     FastPreferenceData<U,I> data)

Description copied from class: ALSFactorizer

User matrix least-squares step.

Specified by:

set_minP in class ALSFactorizer<U,I>

Parameters:

p - user matrix

q - item matrix

data - preference data

set_minQ

public void set_minQ(cern.colt.matrix.impl.DenseDoubleMatrix2D q,
                     cern.colt.matrix.impl.DenseDoubleMatrix2D p,
                     FastPreferenceData<U,I> data)

Description copied from class: ALSFactorizer

Item matrix least-squares step.

Specified by:

set_minQ in class ALSFactorizer<U,I>

Parameters:

q - item matrix

p - user matrix

data - preference data