include/dai/trwbp.h

   1 /*  This file is part of libDAI - http://www.libdai.org/
   2  *
   3  *  Copyright (c) 2006-2011, The libDAI authors. All rights reserved.
   4  *
   5  *  Use of this source code is governed by a BSD-style license that can be found in the LICENSE file.
   6  */
   7
   8
   9 /// \file
  10 /// \brief Defines class TRWBP, which implements Tree-Reweighted Belief Propagation
  11
  12
  13 #ifndef __defined_libdai_trwbp_h
  14 #define __defined_libdai_trwbp_h
  15
  16
  17 #include <string>
  18 #include <dai/daialg.h>
  19 #include <dai/factorgraph.h>
  20 #include <dai/properties.h>
  21 #include <dai/enum.h>
  22 #include <dai/bp.h>
  23
  24
  25 namespace dai {
  26
  27
  28 /// Approximate inference algorithm "Tree-Reweighted Belief Propagation" [\ref WJW03]
  29 /** The Tree-Reweighted Belief Propagation algorithm is like Belief
  30  *  Propagation, but associates each factor with a scale parameter.
  31  *  which controls the divergence measure being minimized.
  32  *
  33  *  The messages \f$m_{I\to i}(x_i)\f$ are passed from factors \f$I\f$ to variables \f$i\f$.
  34  *  The update equation is given by:
  35  *    \f[ m_{I\to i}(x_i) \propto \sum_{x_{N_I\setminus\{i\}}} f_I(x_I)^{1/c_I} \prod_{j\in N_I\setminus\{i\}} m_{I\to j}^{c_I-1} \prod_{J\in N_j\setminus\{I\}} m_{J\to j}^{c_J} \f]
  36  *  After convergence, the variable beliefs are calculated by:
  37  *    \f[ b_i(x_i) \propto \prod_{I\in N_i} m_{I\to i}^{c_I} \f]
  38  *  and the factor beliefs are calculated by:
  39  *    \f[ b_I(x_I) \propto f_I(x_I)^{1/c_I} \prod_{j \in N_I} m_{I\to j}^{c_I-1} \prod_{J\in N_j\setminus\{I\}} m_{J\to j}^{c_J} \f]
  40  *  The logarithm of the partition sum is approximated by:
  41  *    \f[ \log Z = \sum_{I} \sum_{x_I} b_I(x_I) \big( \log f_I(x_I) - c_I \log b_I(x_I) \big) + \sum_{i} (c_i - 1) \sum_{x_i} b_i(x_i) \log b_i(x_i) \f]
  42  *  where the variable weights are defined as
  43  *    \f[ c_i := \sum_{I \in N_i} c_I \f]
  44  *
  45  *  \note TRWBP is actually equivalent to FBP
  46  *  \todo Merge code of FBP and TRWBP
  47  */
  48 class TRWBP : public BP {
  49     protected:
  50         /// "Edge weights" (indexed by factor ID)
  51         /** In [\ref WJW03], only unary or pairwise factors are considered.
  52          *  Here we are more general by having a weight for each factor in the
  53          *  factor graph. If unary factors have weight 1, and higher-order factors
  54          *  are absent, then we have the special case considered in [\ref WJW03].
  55          */
  56         std::vector<Real> _weight;
  57
  58     public:
  59         /// Size of sample of trees used to set the weights
  60         /** \todo See if there is a way to wrap TRWBP::nrtrees in a props struct
  61          *  together with the other properties currently in TRWBP::props
  62          *  (without copying al lot of BP code literally)
  63          */
  64         size_t nrtrees;
  65
  66     public:
  67     /// \name Constructors/destructors
  68     //@{
  69         /// Default constructor
  70         TRWBP() : BP(), _weight() {}
  71
  72         /// Construct from FactorGraph \a fg and PropertySet \a opts.
  73         /** There is an additional property "nrtrees" which allows to specify the
  74          *  number of random spanning trees used to set the scale parameters.
  75          *  \param fg Factor graph.
  76          *  \param opts Parameters @see BP::Properties.
  77          */
  78         TRWBP( const FactorGraph &fg, const PropertySet &opts ) : BP(fg, opts), _weight() {
  79             setProperties( opts );
  80             construct();
  81         }
  82     //@}
  83
  84     /// \name General InfAlg interface
  85     //@{
  86         virtual TRWBP* clone() const { return new TRWBP(*this); }
  87         virtual std::string name() const { return "TRWBP"; }
  88         virtual Real logZ() const;
  89         virtual void setProperties( const PropertySet &opts );
  90         virtual PropertySet getProperties() const;
  91         virtual std::string printProperties() const;
  92     //@}
  93
  94     /// \name TRWBP accessors/mutators for scale parameters
  95     //@{
  96         /// Returns weight corresponding to the \a I 'th factor
  97         Real Weight( size_t I ) const { return _weight[I]; }
  98
  99         /// Returns constant reference to vector of all weights
 100         const std::vector<Real>& Weights() const { return _weight; }
 101
 102         /// Sets the weight of the \a I 'th factor to \a c
 103         void setWeight( size_t I, Real c ) { _weight[I] = c; }
 104
 105         /// Sets the weights of all factors simultaenously
 106         /** \note Faster than calling setWeight(size_t,Real) for each factor
 107          */
 108         void setWeights( const std::vector<Real> &c ) { _weight = c; }
 109
 110         /// Increases weights corresponding to pairwise factors in \a tree with 1
 111         void addTreeToWeights( const RootedTree &tree );
 112
 113         /// Samples weights from a sample of \a nrTrees random spanning trees
 114         void sampleWeights( size_t nrTrees );
 115
 116     protected:
 117         /// Calculate the product of factor \a I and the incoming messages
 118         /** If \a without_i == \c true, the message coming from variable \a i is omitted from the product
 119          *  \note This function is used by calcNewMessage() and calcBeliefF()
 120          */
 121         virtual Prob calcIncomingMessageProduct( size_t I, bool without_i, size_t i ) const;
 122
 123         /// Calculates unnormalized belief of variable \a i
 124         virtual void calcBeliefV( size_t i, Prob &p ) const;
 125
 126         // Calculates unnormalized belief of factor \a I
 127         virtual void calcBeliefF( size_t I, Prob &p ) const {
 128             p = calcIncomingMessageProduct( I, false, 0 );
 129         }
 130
 131         // Helper function for constructors
 132         virtual void construct();
 133 };
 134
 135
 136 } // end of namespace dai
 137
 138
 139 #endif