mlpack::svd::SVDPlusPlus< OptimizerType > Class Template Reference

SVD++ is a matrix decomposition tenique used in collaborative filtering. More...

#include <svdplusplus.hpp>

Public Member Functions
	SVDPlusPlus (const size_t iterations=10, const double alpha=0.001, const double lambda=0.1)
	Constructor of SVDPlusPlus. More...
void	Apply (const arma::mat &data, const arma::mat &implicitData, const size_t rank, arma::mat &u, arma::mat &v, arma::vec &p, arma::vec &q, arma::mat &y)
	Trains the model and obtains user/item matrices, user/item bias, and item implicit matrix. More...
void	Apply (const arma::mat &data, const size_t rank, arma::mat &u, arma::mat &v, arma::vec &p, arma::vec &q, arma::mat &y)
	Trains the model and obtains user/item matrices, user/item bias, and item implicit matrix. More...

Static Public Member Functions
static void	CleanData (const arma::mat &implicitData, arma::sp_mat &cleanedData, const arma::mat &data)
	Converts the User, Item matrix of implicit data to Item-User Table.

Detailed Description

template<typename OptimizerType = ens::StandardSGD>
class mlpack::svd::SVDPlusPlus< OptimizerType >

SVD++ is a matrix decomposition tenique used in collaborative filtering.

SVD++ is similar to BiasSVD, but it is a more expressive model because SVD++ also models implicit feedback. SVD++ outputs user/item latent vectors, user/item bias, and item vectors with regard to implicit feedback. Parameters are optmized by Stochastic Gradient Desent(SGD). The updates also penalize the learning of large feature values by means of regularization.

For more information, see the following paper:

@inproceedings{koren2008factorization,
title={Factorization meets the neighborhood: a multifaceted collaborative
       filtering model},
author={Koren, Yehuda},
booktitle={Proceedings of the 14th ACM SIGKDD international conference on
           Knowledge discovery and data mining},
pages={426--434},
year={2008},
organization={ACM}
}

An example of how to use the interface is shown below:

arma::mat data; // Rating data in the form of coordinate list.

// Implicit feedback data in the form of coordinate list.
arma::mat implicitData;

const size_t rank = 10; // Rank used for the decomposition.
const size_t iterations = 10; // Number of iterations used for optimization.

const double alpha = 0.001 // Learning rate for the SGD optimizer.
const double lambda = 0.1 // Regularization parameter for the optimization.

// Make a SVD++ object.
SVDPlusPlus<> svdPP(iterations, alpha, lambda);

arma::mat u, v; // Item and User matrices.
arma::vec p, q; // Item and User bias.
arma::mat y;    // Item matrix with respect to implicit feedback.

// Use the Apply() method to get a factorization.
svdPP.Apply(data, implicitData, rank, u, v, p, q, y);

Constructor & Destructor Documentation

SVDPlusPlus()

template<typename OptimizerType >

mlpack::svd::SVDPlusPlus< OptimizerType >::SVDPlusPlus	(	const size_t	iterations = 10,
		const double	alpha = 0.001,
		const double	lambda = 0.1
	)

Constructor of SVDPlusPlus.

By default SGD optimizer is used in SVDPlusPlus. The optimizer uses a template specialization of Optimize().

Parameters

iterations	Number of optimization iterations.
alpha	Learning rate for the SGD optimizer.
lambda	Regularization parameter for the optimization.

Member Function Documentation

Apply() [1/2]

template<typename OptimizerType >

void mlpack::svd::SVDPlusPlus< OptimizerType >::Apply	(	const arma::mat &	data,
		const arma::mat &	implicitData,
		const size_t	rank,
		arma::mat &	u,
		arma::mat &	v,
		arma::vec &	p,
		arma::vec &	q,
		arma::mat &	y
	)

Trains the model and obtains user/item matrices, user/item bias, and item implicit matrix.

Parameters

data	Rating data matrix.
implicitData	Implicit feedback.
rank	Rank parameter to be used for optimization.
u	Item matrix obtained on decomposition.
v	User matrix obtained on decomposition.
p	Item bias.
q	User bias.
y	Item matrix with respect to implicit feedback.

Apply() [2/2]

template<typename OptimizerType >

void mlpack::svd::SVDPlusPlus< OptimizerType >::Apply	(	const arma::mat &	data,
		const size_t	rank,
		arma::mat &	u,
		arma::mat &	v,
		arma::vec &	p,
		arma::vec &	q,
		arma::mat &	y
	)

Trains the model and obtains user/item matrices, user/item bias, and item implicit matrix.

Whether a user rates an item is used as implicit feedback.

Parameters

data	Rating data matrix.
rank	Rank parameter to be used for optimization.
u	Item matrix obtained on decomposition.
v	User matrix obtained on decomposition.
p	Item bias.
q	User bias.
y	Item matrix with respect to implicit feedback. Each column is a latent vector of an item with respect to implicit feedback.

---

The documentation for this class was generated from the following files:

• src/mlpack/methods/svdplusplus/svdplusplus.hpp
• src/mlpack/methods/svdplusplus/svdplusplus_impl.hpp