scoreBDeu_tpl.h

Go to the documentation of this file.

#ifndef DOXYGEN_SHOULD_SKIP_THIS

#  include <agrum/learning/scores_and_tests/scoreBDeu.h>
#  include <sstream>

namespace gum {

  namespace learning {

    template < template < typename > class ALLOC >
    INLINE ScoreBDeu< ALLOC >::ScoreBDeu(
       const DBRowGeneratorParser< ALLOC >& parser,
       const Apriori< ALLOC >&              apriori,
       const std::vector< std::pair< std::size_t, std::size_t >,
                          ALLOC< std::pair< std::size_t, std::size_t > > >& ranges,
       const Bijection< NodeId, std::size_t, ALLOC< std::size_t > >&
                                                          nodeId2columns,
       const typename ScoreBDeu< ALLOC >::allocator_type& alloc) :
        Score< ALLOC >(parser, apriori, ranges, nodeId2columns, alloc),
        __internal_apriori(parser.database(), nodeId2columns) {
      GUM_CONSTRUCTOR(ScoreBDeu);
    }


    template < template < typename > class ALLOC >
    INLINE ScoreBDeu< ALLOC >::ScoreBDeu(
       const DBRowGeneratorParser< ALLOC >& parser,
       const Apriori< ALLOC >&              apriori,
       const Bijection< NodeId, std::size_t, ALLOC< std::size_t > >&
                                                          nodeId2columns,
       const typename ScoreBDeu< ALLOC >::allocator_type& alloc) :
        Score< ALLOC >(parser, apriori, nodeId2columns, alloc),
        __internal_apriori(parser.database(), nodeId2columns) {
      GUM_CONSTRUCTOR(ScoreBDeu);
    }


    template < template < typename > class ALLOC >
    INLINE ScoreBDeu< ALLOC >::ScoreBDeu(
       const ScoreBDeu< ALLOC >&                          from,
       const typename ScoreBDeu< ALLOC >::allocator_type& alloc) :
        Score< ALLOC >(from, alloc),
        __internal_apriori(from.__internal_apriori, alloc),
        __gammalog2(from.__gammalog2) {
      GUM_CONS_CPY(ScoreBDeu);
    }


    template < template < typename > class ALLOC >
    INLINE ScoreBDeu< ALLOC >::ScoreBDeu(const ScoreBDeu< ALLOC >& from) :
        ScoreBDeu< ALLOC >(from, from.getAllocator()) {}


    template < template < typename > class ALLOC >
    INLINE ScoreBDeu< ALLOC >::ScoreBDeu(
       ScoreBDeu< ALLOC >&&                               from,
       const typename ScoreBDeu< ALLOC >::allocator_type& alloc) :
        Score< ALLOC >(std::move(from), alloc),
        __internal_apriori(std::move(from.__internal_apriori), alloc),
        __gammalog2(std::move(from.__gammalog2)) {
      GUM_CONS_MOV(ScoreBDeu);
    }


    template < template < typename > class ALLOC >
    INLINE ScoreBDeu< ALLOC >::ScoreBDeu(ScoreBDeu< ALLOC >&& from) :
        ScoreBDeu< ALLOC >(std::move(from), from.getAllocator()) {}


    template < template < typename > class ALLOC >
    ScoreBDeu< ALLOC >* ScoreBDeu< ALLOC >::clone(
       const typename ScoreBDeu< ALLOC >::allocator_type& alloc) const {
      ALLOC< ScoreBDeu< ALLOC > > allocator(alloc);
      ScoreBDeu< ALLOC >*         new_score = allocator.allocate(1);
      try {
        allocator.construct(new_score, *this, alloc);
      } catch (...) {
        allocator.deallocate(new_score, 1);
        throw;
      }

      return new_score;
    }


    template < template < typename > class ALLOC >
    ScoreBDeu< ALLOC >* ScoreBDeu< ALLOC >::clone() const {
      return clone(this->getAllocator());
    }


    template < template < typename > class ALLOC >
    ScoreBDeu< ALLOC >::~ScoreBDeu() {
      GUM_DESTRUCTOR(ScoreBDeu);
    }


    template < template < typename > class ALLOC >
    ScoreBDeu< ALLOC >& ScoreBDeu< ALLOC >::
                        operator=(const ScoreBDeu< ALLOC >& from) {
      if (this != &from) {
        Score< ALLOC >::operator=(from);
        __internal_apriori = from.__internal_apriori;
      }
      return *this;
    }


    template < template < typename > class ALLOC >
    ScoreBDeu< ALLOC >& ScoreBDeu< ALLOC >::operator=(ScoreBDeu< ALLOC >&& from) {
      if (this != &from) {
        Score< ALLOC >::operator=(std::move(from));
        __internal_apriori = std::move(from.__internal_apriori);
      }
      return *this;
    }


    template < template < typename > class ALLOC >
    std::string
       ScoreBDeu< ALLOC >::isAprioriCompatible(const std::string& apriori_type,
                                               double             weight) {
      // check that the apriori is compatible with the score
      if (apriori_type == AprioriNoAprioriType::type) { return ""; }

      if (weight == 0.0) {
        return "The apriori is currently compatible with the BDeu score but "
               "if you change the weight, it will become incompatible.";
      }

      // known incompatible aprioris
      if ((apriori_type == AprioriDirichletType::type)
          || (apriori_type == AprioriSmoothingType::type)) {
        return "The BDeu score already contains a different 'implicit' apriori. "
               "Therefore, the learning will probably be biased.";
      }

      // apriori types unsupported by the type checker
      std::stringstream msg;
      msg << "The apriori '" << apriori_type
          << "' is not yet supported by method isAprioriCompatible os Score BDeu";
      return msg.str();
    }


    template < template < typename > class ALLOC >
    INLINE std::string
           ScoreBDeu< ALLOC >::isAprioriCompatible(const Apriori< ALLOC >& apriori) {
      return isAprioriCompatible(apriori.getType(), apriori.weight());
    }


    template < template < typename > class ALLOC >
    INLINE std::string ScoreBDeu< ALLOC >::isAprioriCompatible() const {
      return isAprioriCompatible(*(this->_apriori));
    }


    template < template < typename > class ALLOC >
    INLINE const Apriori< ALLOC >& ScoreBDeu< ALLOC >::internalApriori() const {
      return __internal_apriori;
    }


    template < template < typename > class ALLOC >
    INLINE void ScoreBDeu< ALLOC >::setEffectiveSampleSize(double ess) {
      if (ess < 0) {
        GUM_ERROR(OutOfBounds,
                  "The effective sample size of the BDeu's "
                  "internal apriori must be positive");
      } else {
        __internal_apriori.setEffectiveSampleSize(ess);
      }
    }


    template < template < typename > class ALLOC >
    double ScoreBDeu< ALLOC >::_score(const IdSet< ALLOC >& idset) {
      // get the counts for all the nodes in the idset and add the apriori
      std::vector< double, ALLOC< double > > N_ijk(
         this->_counter.counts(idset, true));
      const std::size_t all_size = N_ijk.size();

      double       score = 0.0;
      const double ess = __internal_apriori.weight();
      const bool   informative_external_apriori = this->_apriori->isInformative();


      // here, we distinguish idsets with conditioning nodes from those
      // without conditioning nodes
      if (idset.hasConditioningSet()) {
        // get the counts for the conditioning nodes
        std::vector< double, ALLOC< double > > N_ij(
           this->_marginalize(idset[0], N_ijk));
        const std::size_t conditioning_size = N_ij.size();
        const double      ess_qi = ess / conditioning_size;
        const double      ess_riqi = ess / all_size;

        if (informative_external_apriori) {
          // the score to compute is that of BD with aprioris
          // N'_ijk + ESS / (r_i * q_i )
          // (the + ESS / (r_i * q_i ) is here to take into account the
          // internal apriori of BDeu)
          std::vector< double, ALLOC< double > > N_prime_ijk(all_size, 0.0);
          this->_apriori->addAllApriori(idset, N_prime_ijk);
          std::vector< double, ALLOC< double > > N_prime_ij(N_ij.size(), 0.0);
          this->_apriori->addConditioningApriori(idset, N_prime_ij);

          // the BDeu score can be computed as follows:
          // sum_j=1^qi [ gammalog2 ( N'_ij + ESS / q_i ) -
          //     gammalog2 ( N_ij + N'_ij + ESS / q_i )
          //     + sum_k=1^ri { gammlog2 ( N_ijk + N'_ijk + ESS / (r_i * q_i ) )
          //     - gammalog2 ( N'_ijk + ESS / (r_i * q_i ) ) } ]
          for (std::size_t j = std::size_t(0); j < conditioning_size; ++j) {
            score += __gammalog2(N_prime_ij[j] + ess_qi)
                     - __gammalog2(N_ij[j] + N_prime_ij[j] + ess_qi);
          }
          for (std::size_t k = std::size_t(0); k < all_size; ++k) {
            score += __gammalog2(N_ijk[k] + N_prime_ijk[k] + ess_riqi)
                     - __gammalog2(N_prime_ijk[k] + ess_riqi);
          }
        } else {
          // the BDeu score can be computed as follows:
          // qi * gammalog2 (ess / qi) - ri * qi * gammalog2 (ess / (ri * qi) )
          // - sum_j=1^qi [ gammalog2 ( N_ij + ess / qi ) ]
          // + sum_j=1^qi sum_k=1^ri log [ gammalog2 ( N_ijk + ess / (ri * qi) )
          // ]
          score = conditioning_size * __gammalog2(ess_qi)
                  - all_size * __gammalog2(ess_riqi);

          for (const auto n_ij : N_ij) {
            score -= __gammalog2(n_ij + ess_qi);
          }
          for (const auto n_ijk : N_ijk) {
            score += __gammalog2(n_ijk + ess_riqi);
          }
        }
      } else {
        // here, there are no conditioning nodes
        const double ess_ri = ess / all_size;

        if (informative_external_apriori) {
          // the score to compute is that of BD with aprioris
          // N'_ijk + ESS / ( ri * qi )
          // (the + ESS / ( ri * qi ) is here to take into account the
          // internal apriori of K2)
          std::vector< double, ALLOC< double > > N_prime_ijk(all_size, 0.0);
          this->_apriori->addAllApriori(idset, N_prime_ijk);

          // the BDeu score can be computed as follows:
          // gammalog2 ( N' + ess ) - gammalog2 ( N + N' + ess )
          // + sum_k=1^ri { gammlog2 ( N_i + N'_i + ESS / ri)
          //                - gammalog2 ( N'_i + ESS / ri ) }
          double N = 0.0;
          double N_prime = 0.0;
          for (std::size_t k = std::size_t(0); k < all_size; ++k) {
            score += __gammalog2(N_ijk[k] + N_prime_ijk[k] + ess_ri)
                     - __gammalog2(N_prime_ijk[k] + ess_ri);
            N += N_ijk[k];
            N_prime += N_prime_ijk[k];
          }
          score += __gammalog2(N_prime + ess) - __gammalog2(N + N_prime + ess);
        } else {
          // the BDeu score can be computed as follows:
          // gammalog2 ( ess ) - ri * gammalog2 ( ess / ri )
          // - gammalog2 ( N + ess )
          // + sum_k=1^ri log [ gammalog2 ( N_ijk + ess / ri ) ]

          score = __gammalog2(ess) - all_size * __gammalog2(ess_ri);
          double N = 0;
          for (const auto n_ijk : N_ijk) {
            score += __gammalog2(n_ijk + ess_ri);
            N += n_ijk;
          }
          score -= __gammalog2(N + ess);
        }
      }

      return score;
    }

  } /* namespace learning */

} /* namespace gum */

#endif /* DOXYGEN_SHOULD_SKIP_THIS */