recordCounter.h

Go to the documentation of this file.

#ifndef GUM_LEARNING_RECORD_COUNTER_H
#define GUM_LEARNING_RECORD_COUNTER_H

#include <vector>
#include <utility>
#include <sstream>
#include <string>

#include <agrum/agrum.h>
#include <agrum/core/bijection.h>
#include <agrum/core/sequence.h>
#include <agrum/core/OMPThreads.h>
#include <agrum/core/threadData.h>
#include <agrum/graphs/DAG.h>
#include <agrum/learning/database/DBRowGeneratorParser.h>
#include <agrum/learning/scores_and_tests/idSet.h>


namespace gum {

  namespace learning {

    template < template < typename > class ALLOC = std::allocator >
    class RecordCounter {
      public:
      using allocator_type = ALLOC< NodeId >;

      // ##########################################################################
      // ##########################################################################


      RecordCounter(
         const DBRowGeneratorParser< ALLOC >& parser,
         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 =
               Bijection< NodeId, std::size_t, ALLOC< std::size_t > >(),
         const allocator_type& alloc = allocator_type());


      RecordCounter(const DBRowGeneratorParser< ALLOC >& parser,
                    const Bijection< NodeId, std::size_t, ALLOC< std::size_t > >&
                       nodeId2columns =
                          Bijection< NodeId, std::size_t, ALLOC< std::size_t > >(),
                    const allocator_type& alloc = allocator_type());

      RecordCounter(const RecordCounter< ALLOC >& from);

      RecordCounter(const RecordCounter< ALLOC >& from,
                    const allocator_type&         alloc);

      RecordCounter(RecordCounter< ALLOC >&& from);

      RecordCounter(RecordCounter< ALLOC >&& from, const allocator_type& alloc);

      virtual RecordCounter< ALLOC >* clone() const;

      virtual RecordCounter< ALLOC >* clone(const allocator_type& alloc) const;

      virtual ~RecordCounter();



      // ##########################################################################
      // ##########################################################################


      RecordCounter< ALLOC >& operator=(const RecordCounter< ALLOC >& from);

      RecordCounter< ALLOC >& operator=(RecordCounter< ALLOC >&& from);



      // ##########################################################################
      // ##########################################################################


      void clear();

      void setMaxNbThreads(const std::size_t nb) const;

      std::size_t nbThreads() const;

      void setMinNbRowsPerThread(const std::size_t nb) const;

      std::size_t minNbRowsPerThread() const;


      const std::vector< double, ALLOC< double > >&
         counts(const IdSet< ALLOC >& ids, const bool check_discrete_vars = false);


      template < template < typename > class XALLOC >
      void setRanges(
         const std::vector< std::pair< std::size_t, std::size_t >,
                            XALLOC< std::pair< std::size_t, std::size_t > > >&
            new_ranges);

      void clearRanges();

      const std::vector< std::pair< std::size_t, std::size_t >,
                         ALLOC< std::pair< std::size_t, std::size_t > > >&
         ranges() const;


      template < typename GUM_SCALAR >
      void setBayesNet(const BayesNet< GUM_SCALAR >& new_bn);

      allocator_type getAllocator() const;


      const Bijection< NodeId, std::size_t, ALLOC< std::size_t > >&
         nodeId2Columns() const;

      const DatabaseTable< ALLOC >& database() const;



#ifndef DOXYGEN_SHOULD_SKIP_THIS

      private:
      // the parsers used by the threads
      std::vector< ThreadData< DBRowGeneratorParser< ALLOC > >,
                   ALLOC< ThreadData< DBRowGeneratorParser< ALLOC > > > >
         __parsers;

      // the set of ranges of the database's rows indices over which the user
      // wishes to perform the countings
      std::vector< std::pair< std::size_t, std::size_t >,
                   ALLOC< std::pair< std::size_t, std::size_t > > >
         __ranges;

      // the ranges actually used by the threads: there is a hopefully clever
      // algorithm that split the rows ranges into another set of ranges that
      // are assigned to the threads. For instance, if the database has 1000
      // rows and there are 10 threads, each one will be assed a set of 100
      // rows. These sets are precisely what are stored in the field below
      mutable std::vector< std::pair< std::size_t, std::size_t >,
                           ALLOC< std::pair< std::size_t, std::size_t > > >
         __thread_ranges;

      // the mapping from the NodeIds of the variables to the indices of the
      // columns in the database
      Bijection< NodeId, std::size_t, ALLOC< std::size_t > > __nodeId2columns;

      // the last database-parsed countings
      std::vector< double, ALLOC< double > > __last_DB_countings;

      // the ids of the nodes for the last database-parsed countings
      IdSet< ALLOC > __last_DB_ids;

      // the last countings deduced from __last_DB_countings
      std::vector< double, ALLOC< double > > __last_nonDB_countings;

      // the ids of the nodes of last countings deduced from __last_DB_countings
      IdSet< ALLOC > __last_nonDB_ids;

      // the maximal number of threads that the record counter can use
      mutable std::size_t __max_nb_threads{
         std::size_t(gum::getMaxNumberOfThreads())};

      // the min number of rows that a thread should process in a
      // multithreading context
      mutable std::size_t __min_nb_rows_per_thread{100};

      // returns a mapping from the nodes ids to the columns of the database
      // for a given sequence of ids. This is especially convenient when
      // __nodeId2columns is empty (which means that there is an identity mapping)
      HashTable< NodeId, std::size_t >
         __getNodeIds2Columns(const IdSet< ALLOC >& ids) const;

      std::vector< double, ALLOC< double > >& __extractFromCountings(
         const IdSet< ALLOC >&                         subset_ids,
         const IdSet< ALLOC >&                         superset_ids,
         const std::vector< double, ALLOC< double > >& superset_vect);

      std::vector< double, ALLOC< double > >&
         __countFromDatabase(const IdSet< ALLOC >& ids);

      void __threadedCount(
         const std::size_t              range_begin,
         const std::size_t              range_end,
         DBRowGeneratorParser< ALLOC >& parser,
         const std::vector< std::pair< std::size_t, std::size_t >,
                            ALLOC< std::pair< std::size_t, std::size_t > > >&
                                                 cols_and_offsets,
         std::vector< double, ALLOC< double > >& countings);


      template < template < typename > class XALLOC >
      void __checkRanges(
         const std::vector< std::pair< std::size_t, std::size_t >,
                            XALLOC< std::pair< std::size_t, std::size_t > > >&
            new_ranges) const;


      void __checkDiscreteVariables(const IdSet< ALLOC >& ids) const;


      void __raiseCheckException(
         const std::vector< std::string, ALLOC< std::string > >& bad_vars) const;

      void __dispatchRangesToThreads();

#endif /* DOXYGEN_SHOULD_SKIP_THIS */
    };

  } /* namespace learning */

} /* namespace gum */

#include <agrum/learning/scores_and_tests/recordCounter_tpl.h>

#endif /* GUM_LEARNING_RECORD_COUNTER_H */