BNDatabaseGenerator_tpl.h

Go to the documentation of this file.

/***************************************************************************
 *   Copyright (C) 2005 by Christophe GONZALES and Pierre-Henri WUILLEMIN  *
 *   {prenom.nom}@lip6.fr                                                  *
 *                                                                         *
 *   This program is free software; you can redistribute it and/or modify  *
 *   it under the terms of the GNU General Public License as published by  *
 *   the Free Software Foundation; either version 2 of the License, or     *
 *   (at your option) any later version.                                   *
 *                                                                         *
 *   This program is distributed in the hope that it wil be useful,        *
 *   but WITHOUT ANY WARRANTY; without even the implied warranty of        *
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the         *
 *   GNU General Public License for more details.                          *
 *                                                                         *
 *   You should have received a copy of the GNU General Public License     *
 *   along with this program; if not, write to the                         *
 *   Free Software Foundation, Inc.,                                       *
 *   59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.             *
 ***************************************************************************/
#include <agrum/learning/database/BNDatabaseGenerator.h>

#include <agrum/core/timer.h>


namespace gum {
  namespace learning {


    template < typename GUM_SCALAR >
    BNDatabaseGenerator< GUM_SCALAR >::BNDatabaseGenerator(
       const BayesNet< GUM_SCALAR >& bn) :
        __bn(bn) {
      // for debugging purposes
      GUM_CONSTRUCTOR(BNDatabaseGenerator);

      // get the node names => they will serve as ids
      NodeId id = 0;
      for (const auto& var : __bn.dag()) {
        auto name = __bn.variable(var).name();
        __names2ids.insert(name, var);
        ++id;
      }
      __nbVars = id;
      __varOrder.resize(__nbVars);
      std::iota(__varOrder.begin(), __varOrder.end(), (Idx)0);
    }

    template < typename GUM_SCALAR >
    BNDatabaseGenerator< GUM_SCALAR >::~BNDatabaseGenerator() {
      GUM_DESTRUCTOR(BNDatabaseGenerator);
    }


    template < typename GUM_SCALAR >
    double BNDatabaseGenerator< GUM_SCALAR >::drawSamples(Size nbSamples) {
      Timer timer;
      int   progress = 0;

      timer.reset();

      if (onProgress.hasListener()) {
        GUM_EMIT2(onProgress, progress, timer.step());
      }
      __database.clear();
      __database.resize(nbSamples);
      for (auto& row : __database) {
        row.resize(__nbVars);
      }
      // get the order in which the nodes will be sampled
      const gum::Sequence< gum::NodeId >& topOrder = __bn.topologicalOrder();
      std::vector< gum::Instantiation >   instantiations;

      // create instantiations in advance
      for (Idx node = 0; node < __nbVars; ++node)
        instantiations.push_back(gum::Instantiation(__bn.cpt(node)));

      // create the random generator
      std::random_device               rd;
      std::mt19937                     gen(rd());
      std::uniform_real_distribution<> distro(0.0, 1.0);

      // perform the sampling
      __log2likelihood = 0;
      const gum::DAG& dag = __bn.dag();
      for (Idx i = 0; i < nbSamples; ++i) {
        if (onProgress.hasListener()) {
          int p = int((i * 100) / nbSamples);
          if (p != progress) {
            progress = p;
            GUM_EMIT2(onProgress, progress, timer.step());
          }
        }
        std::vector< Idx >& sample = __database.at(i);
        for (Idx j = 0; j < __nbVars; ++j) {
          const gum::NodeId node = topOrder[j];
          const auto&       var = __bn.variable(node);
          const auto&       cpt = __bn.cpt(node);

          gum::Instantiation& inst = instantiations[node];
          for (auto par : dag.parents(node))
            inst.chgVal(__bn.variable(par), sample.at(par));

          const double nb = distro(gen);
          double       cumul = 0.0;
          for (inst.chgVal(var, 0); !inst.end(); inst.incVar(var)) {
            cumul += cpt[inst];
            if (cumul >= nb) break;
          }

          if (inst.end()) inst.chgVal(var, var.domainSize() - 1);
          sample.at(node) = inst.val(var);

          __log2likelihood += std::log2(__bn.cpt(node)[inst]);
        }
      }

      __drawnSamples = true;

      if (onProgress.hasListener()) {
        std::stringstream ss;
        ss << "Database of size " << nbSamples << " generated in " << timer.step()
           << " seconds. Log2likelihood : " << __log2likelihood;
        GUM_EMIT1(onStop, ss.str());
      }

      return __log2likelihood;
    }

    template < typename GUM_SCALAR >
    void BNDatabaseGenerator< GUM_SCALAR >::toCSV(const std::string& csvFileURL,
                                                  bool               useLabels,
                                                  bool               append,
                                                  std::string        csvSeparator,
                                                  bool checkOnAppend) const {
      if (!__drawnSamples) {
        GUM_ERROR(OperationNotAllowed, "drawSamples() must be called first.");
      }

      if (csvSeparator.find("\n") != std::string::npos) {
        GUM_ERROR(InvalidArgument,
                  "csvSeparator must not contain end-line characters");
      }

      bool includeHeader = true;
      if (append) {
        std::ifstream csvFile(csvFileURL);
        if (csvFile) {
          auto varOrder = __varOrderFromCSV(csvFile, csvSeparator);
          if (checkOnAppend && varOrder != __varOrder) {
            GUM_ERROR(
               OperationNotAllowed,
               "Inconsistent variable order in csvFile when appending. You "
               "can use setVarOrderFromCSV(url) function to get the right "
               "order. You could also set parameter checkOnAppend=false if you "
               "know what you are doing.");
          }
          includeHeader = false;
        }
        csvFile.close();
      }


      auto ofstreamFlag = append ? std::ofstream::app : std::ofstream::out;

      std::ofstream os(csvFileURL, ofstreamFlag);
      bool          firstCol = true;
      if (includeHeader) {
        for (const auto& i : __varOrder) {
          if (firstCol) {
            firstCol = false;
          } else {
            os << csvSeparator;
          }
          os << __bn.variable(i).name();
        }
      }
      os << std::endl;

      bool firstRow = true;
      for (const auto& row : __database) {
        if (firstRow) {
          firstRow = false;
        } else {
          os << std::endl;
        }
        firstCol = true;
        for (const auto& i : __varOrder) {
          if (firstCol) {
            firstCol = false;
          } else {
            os << csvSeparator;
          }
          if (useLabels) {
            os << __bn.variable(i).label(row.at(i));
          } else {
            os << row[i];
          }
        }
      }

      os.close();
    }

    template < typename GUM_SCALAR >
    DatabaseTable<>
       BNDatabaseGenerator< GUM_SCALAR >::toDatabaseTable(bool useLabels) const {
      if (!__drawnSamples)
        GUM_ERROR(OperationNotAllowed, "proceed() must be called first.");

      DatabaseTable<>            db;
      std::vector< std::string > varNames;
      varNames.reserve(__nbVars);
      for (const auto& i : __varOrder) {
        varNames.push_back(__names2ids.first(i));
      }

      // create the translators
      for (std::size_t i = 0; i < __nbVars; ++i) {
        const Variable& var = __bn.variable(__varOrder[i]);
        db.insertTranslator(var, i);
      }


      // db.setVariableNames(varNames);
      // db.setVariableNames(varOrderNames());

      if (useLabels) {
        std::vector< std::string > xrow(__nbVars);
        for (const auto& row : __database) {
          for (Idx i = 0; i < __nbVars; ++i) {
            Idx j = __varOrder.at(i);
            xrow[i] = __bn.variable(j).label(row.at(j));
          }
          db.insertRow(xrow);
        }
      } else {
        std::vector< DBTranslatedValueType > translatorType(__nbVars);
        for (std::size_t i = 0; i < __nbVars; ++i) {
          translatorType[i] = db.translator(i).getValType();
        }
        DBRow< DBTranslatedValue > xrow(__nbVars);
        const auto xmiss = gum::learning::DatabaseTable<>::IsMissing::False;
        for (const auto& row : __database) {
          for (Idx i = 0; i < __nbVars; ++i) {
            Idx j = __varOrder.at(i);

            if (translatorType[i] == DBTranslatedValueType::DISCRETE)
              xrow[i].discr_val = std::size_t(row.at(j));
            else
              xrow[i].cont_val = float(row.at(j));
          }
        }
        db.insertRow(xrow, xmiss);
      }

      return db;
    }


    template < typename GUM_SCALAR >
    std::vector< std::vector< Idx > >
       BNDatabaseGenerator< GUM_SCALAR >::database() const {
      if (!__drawnSamples)
        GUM_ERROR(OperationNotAllowed, "drawSamples() must be called first.");

      auto db(__database);
      for (Idx i = 0; i < __database.size(); ++i) {
        for (Idx j = 0; j < __nbVars; ++j) {
          db.at(i).at(j) = (Idx)__database.at(i).at(__varOrder.at(j));
        }
      }
      return db;
    }

    template < typename GUM_SCALAR >
    void BNDatabaseGenerator< GUM_SCALAR >::setVarOrder(
       const std::vector< Idx >& varOrder) {
      if (varOrder.size() != __nbVars) {
        GUM_ERROR(FatalError,
                  "varOrder's size must be equal to the number of variables");
      }
      std::vector< bool > usedVars(__nbVars, false);
      for (const auto& i : varOrder) {
        if (i >= __nbVars) {
          GUM_ERROR(FatalError, "varOrder contains invalid variables");
        }
        if (usedVars.at(i))
          GUM_ERROR(FatalError, "varOrder must not have repeated variables");
        usedVars.at(i) = true;
      }

      if (std::find(usedVars.begin(), usedVars.end(), false) != usedVars.end()) {
        GUM_ERROR(FatalError, "varOrder must contain all variables");
      }

      __varOrder = varOrder;
    }

    template < typename GUM_SCALAR >
    void BNDatabaseGenerator< GUM_SCALAR >::setVarOrder(
       const std::vector< std::string >& varOrder) {
      std::vector< Idx > varOrderIdx;
      varOrderIdx.reserve(varOrder.size());
      for (const auto& vname : varOrder) {
        varOrderIdx.push_back(__names2ids.second(vname));
      }
      setVarOrder(varOrderIdx);
    }

    template < typename GUM_SCALAR >
    void BNDatabaseGenerator< GUM_SCALAR >::setVarOrderFromCSV(
       const std::string& csvFileURL, const std::string& csvSeparator) {
      setVarOrder(__varOrderFromCSV(csvFileURL, csvSeparator));
    }

    template < typename GUM_SCALAR >
    void BNDatabaseGenerator< GUM_SCALAR >::setTopologicalVarOrder() {
      std::vector< Idx > varOrder;
      varOrder.reserve(__nbVars);
      for (const auto& v : __bn.topologicalOrder()) {
        varOrder.push_back(v);
      }
      setVarOrder(varOrder);
    }

    template < typename GUM_SCALAR >
    void BNDatabaseGenerator< GUM_SCALAR >::setAntiTopologicalVarOrder() {
      std::vector< Idx > varOrder;
      varOrder.reserve(__nbVars);
      for (const auto& v : __bn.topologicalOrder()) {
        varOrder.push_back(v);
      }
      std::reverse(varOrder.begin(), varOrder.end());
      setVarOrder(varOrder);
    }

    template < typename GUM_SCALAR >
    void BNDatabaseGenerator< GUM_SCALAR >::setRandomVarOrder() {
      std::vector< std::string > varOrder;
      varOrder.reserve(__bn.size());
      for (const auto& var : __bn.dag()) {
        varOrder.push_back(__bn.variable(var).name());
      }
      std::random_device rd;
      std::mt19937       g(rd());
      std::shuffle(varOrder.begin(), varOrder.end(), g);
      setVarOrder(varOrder);
    }


    template < typename GUM_SCALAR >
    std::vector< Idx > BNDatabaseGenerator< GUM_SCALAR >::varOrder() const {
      return __varOrder;
    }

    template < typename GUM_SCALAR >
    std::vector< std::string >
       BNDatabaseGenerator< GUM_SCALAR >::varOrderNames() const {
      std::vector< std::string > varNames;
      varNames.reserve(__nbVars);
      for (const auto& i : __varOrder) {
        varNames.push_back(__names2ids.first(i));
      }

      return varNames;
    }

    template < typename GUM_SCALAR >
    double BNDatabaseGenerator< GUM_SCALAR >::log2likelihood() const {
      if (!__drawnSamples) {
        GUM_ERROR(OperationNotAllowed, "drawSamples() must be called first.");
      }
      return __log2likelihood;
    }

    template < typename GUM_SCALAR >
    std::vector< Idx > BNDatabaseGenerator< GUM_SCALAR >::__varOrderFromCSV(
       const std::string& csvFileURL, const std::string& csvSeparator) const {
      std::ifstream      csvFile(csvFileURL);
      std::vector< Idx > varOrder;
      if (csvFile) {
        varOrder = __varOrderFromCSV(csvFile, csvSeparator);
        csvFile.close();
      } else {
        GUM_ERROR(NotFound, "csvFileURL does not exist");
      }

      return varOrder;
    }

    template < typename GUM_SCALAR >
    std::vector< Idx > BNDatabaseGenerator< GUM_SCALAR >::__varOrderFromCSV(
       std::ifstream& csvFile, const std::string& csvSeparator) const {
      std::string                line;
      std::vector< std::string > header_found;
      header_found.reserve(__nbVars);
      while (std::getline(csvFile, line)) {
        std::size_t i = 0;
        auto        pos = line.find(csvSeparator);
        while (pos != std::string::npos) {
          header_found.push_back(line.substr(i, pos - i));
          pos += csvSeparator.length();
          i = pos;
          pos = line.find(csvSeparator, pos);

          if (pos == std::string::npos)
            header_found.push_back(line.substr(i, line.length()));
        }
        break;
      }

      std::vector< Size > varOrder;
      varOrder.reserve(__nbVars);

      for (const auto& hf : header_found) {
        varOrder.push_back(__names2ids.second(hf));
      }

      return varOrder;
    }
  } /* namespace learning */
} /* namespace gum */