svd_complete_method.hpp
Go to the documentation of this file.
1 
14 #ifndef MLPACK_METHODS_CF_DECOMPOSITION_POLICIES_SVD_COMPLETE_METHOD_HPP
15 #define MLPACK_METHODS_CF_DECOMPOSITION_POLICIES_SVD_COMPLETE_METHOD_HPP
16 
17 #include <mlpack/prereqs.hpp>
22 
23 namespace mlpack {
24 namespace cf {
25 
45 {
46  public:
59  template<typename MatType>
60  void Apply(const MatType& /* data */,
61  const arma::sp_mat& cleanedData,
62  const size_t rank,
63  const size_t maxIterations,
64  const double minResidue,
65  const bool mit)
66  {
67  if (mit)
68  {
69  amf::MaxIterationTermination iter(maxIterations);
70 
71  // Do singular value decomposition using complete incremental method
72  // using cleaned data in form of sparse matrix.
75 
76  svdci.Apply(cleanedData, rank, w, h);
77  }
78  else
79  {
80  amf::SimpleResidueTermination srt(minResidue, maxIterations);
81 
82  // Do singular value decomposition using complete incremental method
83  // using cleaned data in form of sparse matrix.
85 
86  svdci.Apply(cleanedData, rank, w, h);
87  }
88  }
89 
96  double GetRating(const size_t user, const size_t item) const
97  {
98  double rating = arma::as_scalar(w.row(item) * h.col(user));
99  return rating;
100  }
101 
108  void GetRatingOfUser(const size_t user, arma::vec& rating) const
109  {
110  rating = w * h.col(user);
111  }
112 
125  template<typename NeighborSearchPolicy>
126  void GetNeighborhood(const arma::Col<size_t>& users,
127  const size_t numUsersForSimilarity,
128  arma::Mat<size_t>& neighborhood,
129  arma::mat& similarities) const
130  {
131  // We want to avoid calculating the full rating matrix, so we will do
132  // nearest neighbor search only on the H matrix, using the observation that
133  // if the rating matrix X = W*H, then d(X.col(i), X.col(j)) = d(W H.col(i),
134  // W H.col(j)). This can be seen as nearest neighbor search on the H
135  // matrix with the Mahalanobis distance where M^{-1} = W^T W. So, we'll
136  // decompose M^{-1} = L L^T (the Cholesky decomposition), and then multiply
137  // H by L^T. Then we can perform nearest neighbor search.
138  arma::mat l = arma::chol(w.t() * w);
139  arma::mat stretchedH = l * h; // Due to the Armadillo API, l is L^T.
140 
141  // Temporarily store feature vector of queried users.
142  arma::mat query(stretchedH.n_rows, users.n_elem);
143  // Select feature vectors of queried users.
144  for (size_t i = 0; i < users.n_elem; i++)
145  query.col(i) = stretchedH.col(users(i));
146 
147  NeighborSearchPolicy neighborSearch(stretchedH);
148  neighborSearch.Search(
149  query, numUsersForSimilarity, neighborhood, similarities);
150  }
151 
153  const arma::mat& W() const { return w; }
155  const arma::mat& H() const { return h; }
156 
160  template<typename Archive>
161  void serialize(Archive& ar, const unsigned int /* version */)
162  {
163  ar & BOOST_SERIALIZATION_NVP(w);
164  ar & BOOST_SERIALIZATION_NVP(h);
165  }
166 
167  private:
169  arma::mat w;
171  arma::mat h;
172 };
173 
174 } // namespace cf
175 } // namespace mlpack
176 
177 #endif
This class implements AMF (alternating matrix factorization) on the given matrix V.
Definition: amf.hpp:78
This initialization rule for AMF simply fills the W and H matrices with uniform random noise in [0...
Definition: random_init.hpp:25
TODO : Merge this template specialized function for sparse matrix using common row_col_iterator.
strip_type.hpp
Definition: add_to_po.hpp:21
const arma::mat & W() const
Get the Item Matrix.
The core includes that mlpack expects; standard C++ includes and Armadillo.
void GetNeighborhood(const arma::Col< size_t > &users, const size_t numUsersForSimilarity, arma::Mat< size_t > &neighborhood, arma::mat &similarities) const
Get the neighborhood and corresponding similarities for a set of users.
This class implements a simple residue-based termination policy.
const arma::mat & H() const
Get the User Matrix.
double GetRating(const size_t user, const size_t item) const
Return predicted rating given user ID and item ID.
Implementation of the SVD complete incremental policy to act as a wrapper when accessing SVD complete...
double Apply(const MatType &V, const size_t r, arma::mat &W, arma::mat &H)
Apply Alternating Matrix Factorization to the provided matrix.
void GetRatingOfUser(const size_t user, arma::vec &rating) const
Get predicted ratings for a user.
void serialize(Archive &ar, const unsigned int)
Serialization.
This termination policy only terminates when the maximum number of iterations has been reached...
void Apply(const MatType &, const arma::sp_mat &cleanedData, const size_t rank, const size_t maxIterations, const double minResidue, const bool mit)
Apply Collaborative Filtering to the provided data set using the SVD complete incremental policy...