IBayesNet_tpl.h

Go to the documentation of this file.

/****************************************************************************
 *   This file is part of the aGrUM/pyAgrum library.                        *
 *                                                                          *
 *   Copyright (c) 2005-2025 by                                             *
 *       - Pierre-Henri WUILLEMIN(_at_LIP6)                                 *
 *       - Christophe GONZALES(_at_AMU)                                     *
 *                                                                          *
 *   The aGrUM/pyAgrum library is free software; you can redistribute it    *
 *   and/or modify it under the terms of either :                           *
 *                                                                          *
 *    - the GNU Lesser General Public License as published by               *
 *      the Free Software Foundation, either version 3 of the License,      *
 *      or (at your option) any later version,                              *
 *    - the MIT license (MIT),                                              *
 *    - or both in dual license, as here.                                   *
 *                                                                          *
 *   (see https://agrum.gitlab.io/articles/dual-licenses-lgplv3mit.html)    *
 *                                                                          *
 *   This aGrUM/pyAgrum library is distributed in the hope that it will be  *
 *   useful, but WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,          *
 *   INCLUDING BUT NOT LIMITED TO THE WARRANTIES MERCHANTABILITY or FITNESS *
 *   FOR A PARTICULAR PURPOSE  AND NONINFRINGEMENT. IN NO EVENT SHALL THE   *
 *   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER *
 *   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,        *
 *   ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR  *
 *   OTHER DEALINGS IN THE SOFTWARE.                                        *
 *                                                                          *
 *   See LICENCES for more details.                                         *
 *                                                                          *
 *   SPDX-FileCopyrightText: Copyright 2005-2025                            *
 *       - Pierre-Henri WUILLEMIN(_at_LIP6)                                 *
 *       - Christophe GONZALES(_at_AMU)                                     *
 *   SPDX-License-Identifier: LGPL-3.0-or-later OR MIT                      *
 *                                                                          *
 *   Contact  : info_at_agrum_dot_org                                       *
 *   homepage : http://agrum.gitlab.io                                      *
 *   gitlab   : https://gitlab.com/agrumery/agrum                           *
 *                                                                          *
 ****************************************************************************/
#pragma once
 

 
#include <cmath>
#include <limits>
 
#include <agrum/base/multidim/aggregators/and.h>
#include <agrum/base/multidim/aggregators/or.h>
#include <agrum/base/multidim/ICIModels/multiDimNoisyAND.h>
#include <agrum/base/multidim/ICIModels/multiDimNoisyORCompound.h>
#include <agrum/base/multidim/ICIModels/multiDimNoisyORNet.h>
#include <agrum/base/multidim/tensor.h>
#include <agrum/BN/generator/simpleCPTGenerator.h>
#include <agrum/BN/IBayesNet.h>
 
namespace gum {
  //                                  IBayesNet
 
  template < typename GUM_SCALAR >
  INLINE IBayesNet< GUM_SCALAR >::IBayesNet() : DAGmodel() {
    GUM_CONSTRUCTOR(IBayesNet)
  }
 
  template < typename GUM_SCALAR >
  INLINE IBayesNet< GUM_SCALAR >::IBayesNet(std::string name) : DAGmodel() {
    GUM_CONSTRUCTOR(IBayesNet)
    this->setProperty("name", name);
  }
 
  template < typename GUM_SCALAR >
  IBayesNet< GUM_SCALAR >::IBayesNet(const IBayesNet< GUM_SCALAR >& source) : DAGmodel(source) {
    GUM_CONS_CPY(IBayesNet)
  }
 
  template < typename GUM_SCALAR >
  IBayesNet< GUM_SCALAR >&
      IBayesNet< GUM_SCALAR >::operator=(const IBayesNet< GUM_SCALAR >& source) {
    if (this != &source) { DAGmodel::operator=(source); }
 
    return *this;
  }
 
  template < typename GUM_SCALAR >
  IBayesNet< GUM_SCALAR >::~IBayesNet() {
    GUM_DESTRUCTOR(IBayesNet)
  }
 
  template < typename GUM_SCALAR >
  Size IBayesNet< GUM_SCALAR >::dim() const {
    Size dim = 0;
 
    for (auto node: nodes()) {
      Size q = 1;
 
      for (auto parent: parents(node))
        q *= variable(parent).domainSize();
 
      dim += (variable(node).domainSize() - 1) * q;
    }
 
    return dim;
  }
 
  template < typename GUM_SCALAR >
  Size IBayesNet< GUM_SCALAR >::maxVarDomainSize() const {
    Size res = 0;
    for (auto node: nodes()) {
      auto v = variable(node).domainSize();
      if (v > res) { res = v; }
    }
    return res;
  }
 
  template < typename GUM_SCALAR >
  GUM_SCALAR IBayesNet< GUM_SCALAR >::minParam() const {
    GUM_SCALAR res = 1.0;
    for (auto node: nodes()) {
      auto v = cpt(node).min();
      if (v < res) { res = v; }
    }
    return res;
  }
 
  template < typename GUM_SCALAR >
  GUM_SCALAR IBayesNet< GUM_SCALAR >::maxParam() const {
    GUM_SCALAR res = 1.0;
    for (auto node: nodes()) {
      auto v = cpt(node).max();
      if (v > res) { res = v; }
    }
    return res;
  }
 
  template < typename GUM_SCALAR >
  GUM_SCALAR IBayesNet< GUM_SCALAR >::minNonZeroParam() const {
    GUM_SCALAR res = 1.0;
    for (auto node: nodes()) {
      auto v = cpt(node).minNonZero();
      if (v < res) { res = v; }
    }
    return res;
  }
 
  template < typename GUM_SCALAR >
  GUM_SCALAR IBayesNet< GUM_SCALAR >::maxNonOneParam() const {
    GUM_SCALAR res = 0.0;
    for (auto node: nodes()) {
      auto v = cpt(node).maxNonOne();
      if (v > res) { res = v; }
    }
    return res;
  }
 
  template < typename GUM_SCALAR >
  INLINE Size IBayesNet< GUM_SCALAR >::memoryFootprint() const {
    Size usedMem = 0;
 
    for (auto node: nodes())
      usedMem += cpt(node).memoryFootprint();
    return usedMem;
  }
 
  template < typename GUM_SCALAR >
  INLINE std::string IBayesNet< GUM_SCALAR >::toString() const {
    std::stringstream s;
    s << "BN{nodes: " << size() << ", arcs: " << dag().sizeArcs() << ", ";
    spaceCplxToStream(s, log10DomainSize(), (int)dim(), memoryFootprint());
    s << "}";
    return s.str();
  }
 
  template < typename GUM_SCALAR >
  std::string IBayesNet< GUM_SCALAR >::toDot() const {
    std::stringstream output;
    output << "digraph \"";
 
    std::string bn_name;
 
    try {
      bn_name = this->property("name");
    } catch (NotFound const&) { bn_name = "no_name"; }
 
    output << bn_name << "\" {" << std::endl;
    output << "  graph [bgcolor=transparent,label=\"" << bn_name << "\"];" << std::endl;
    output << "  node [style=filled fillcolor=\"#ffffaa\"];" << std::endl << std::endl;
 
    for (auto node: nodes())
      output << "\"" << variable(node).name() << "\" [comment=\"" << node << ":"
             << variable(node).toStringWithDescription() << "\"];" << std::endl;
 
    output << std::endl;
 
    std::string tab = "  ";
 
    for (auto node: nodes()) {
      if (children(node).size() > 0) {
        for (auto child: children(node)) {
          output << tab << "\"" << variable(node).name() << "\" -> " << "\""
                 << variable(child).name() << "\";" << std::endl;
        }
      } else if (parents(node).size() == 0) {
        output << tab << "\"" << variable(node).name() << "\";" << std::endl;
      }
    }
 
    output << "}" << std::endl;
 
    return output.str();
  }

  template < typename GUM_SCALAR >
  GUM_SCALAR IBayesNet< GUM_SCALAR >::jointProbability(const Instantiation& i) const {
    auto value = (GUM_SCALAR)1.0;
 
    GUM_SCALAR tmp;
 
    for (auto node: nodes()) {
      if ((tmp = cpt(node)[i]) == (GUM_SCALAR)0) { return (GUM_SCALAR)0; }
 
      value *= tmp;
    }
 
    return value;
  }

  template < typename GUM_SCALAR >
  GUM_SCALAR IBayesNet< GUM_SCALAR >::log2JointProbability(const Instantiation& i) const {
    auto value = (GUM_SCALAR)0.0;
 
    GUM_SCALAR tmp;
 
    for (auto node: nodes()) {
      if ((tmp = cpt(node)[i]) == (GUM_SCALAR)0) {
        return (GUM_SCALAR)(-std::numeric_limits< double >::infinity());
      }
 
      value += std::log2(cpt(node)[i]);
    }
 
    return value;
  }
 
  template < typename GUM_SCALAR >
  bool IBayesNet< GUM_SCALAR >::operator==(const IBayesNet& from) const {
    if (size() != from.size()) { return false; }
 
    if (sizeArcs() != from.sizeArcs()) { return false; }
 
    for (auto node: nodes()) {
      try {
        const auto& v1 = variable(node);
        const auto& v2 = from.variableFromName(variable(node).name());
        if (v1 != v2) { return false; }
      } catch (NotFound const&) {
        // a name is not found in from
        return false;
      }
    }
 
    for (auto node: nodes()) {
      NodeId fromnode = from.idFromName(variable(node).name());
 
      if (cpt(node).nbrDim() != from.cpt(fromnode).nbrDim()) { return false; }
 
      if (cpt(node).domainSize() != from.cpt(fromnode).domainSize()) { return false; }
 
      for (Idx i = 0; i < cpt(node).nbrDim(); ++i) {
        if (!from.cpt(fromnode).contains(from.variableFromName(cpt(node).variable(i).name()))) {
          return false;
        }
      }
 
      Instantiation i(cpt(node));
      Instantiation j(from.cpt(fromnode));
 
      AlmostDifferent< GUM_SCALAR > cmp;
      for (i.setFirst(); !i.end(); i.inc()) {
        for (Idx indice = 0; indice < cpt(node).nbrDim(); ++indice) {
          const DiscreteVariable* p = &(i.variable(indice));
          j.chgVal(j.pos(from.variableFromName(p->name())), i.val(*p));
        }
 
        if (cmp(cpt(node).get(i), from.cpt(fromnode).get(j))) { return false; }
      }
    }
 
    return true;
  }
 
  template < typename GUM_SCALAR >
  bool IBayesNet< GUM_SCALAR >::operator!=(const IBayesNet& from) const {
    return !this->operator==(from);
  }
 
  template < typename GUM_SCALAR >
  INLINE std::ostream& operator<<(std::ostream& output, const IBayesNet< GUM_SCALAR >& bn) {
    output << bn.toString();
    return output;
  }
 
  template < typename GUM_SCALAR >
  std::vector< std::string > IBayesNet< GUM_SCALAR >::check() const {
    std::vector< std::string > comments;
 
    const double epsilon       = 1e-8;
    const double error_epsilon = 1e-1;
 
    // CHECKING domain
    for (const auto i: nodes())
      if (variable(i).domainSize() < 2) {
        std::stringstream s;
        s << "Variable " << variable(i).name() << ": not consistent (domainSize=1).";
        comments.push_back(s.str());
      }
 
    // CHECKING parameters are probabilities
    // >0
    for (const auto i: nodes()) {
      const auto [amin, minval] = cpt(i).argmin();
      if (minval < (GUM_SCALAR)0.0) {
        std::stringstream s;
        s << "Variable " << variable(i).name() << " : P(" << *(amin.begin()) << ") < 0.0";
        comments.push_back(s.str());
      }
    }
    // <1
    for (const auto i: nodes()) {
      const auto [amax, maxval] = cpt(i).argmax();
      if (maxval > (GUM_SCALAR)1.0) {
        std::stringstream s;
        s << "Variable " << variable(i).name() << " : P(" << *(amax.begin()) << ") > 1.0";
        comments.push_back(s.str());
      }
    }
 
    // CHECKING distributions sum to 1
    for (const auto i: nodes()) {
      const auto p              = cpt(i).sumOut({&variable(i)});
      const auto [amin, minval] = p.argmin();
      if (minval < (GUM_SCALAR)(1.0 - epsilon)) {
        std::stringstream s;
        s << "Variable " << variable(i).name() << " : ";
        if (!parents(i).empty()) s << "with (at least) parents " << *(amin.begin()) << ", ";
        s << "the CPT sum to less than 1";
        if (minval > (GUM_SCALAR)(1.0 - error_epsilon)) s << " (normalization problem ?)";
        s << ".";
        comments.push_back(s.str());
        continue;
      }
      const auto [amax, maxval] = p.argmax();
      if (maxval > (GUM_SCALAR)(1.0 + epsilon)) {
        std::stringstream s;
        s << "Variable " << variable(i).name() << " : ";
        if (!parents(i).empty()) s << "with (at least) parents " << *(amax.begin()) << ", ";
        s << "the CPT sum to more than 1";
        if (maxval < (GUM_SCALAR)(1.0 + error_epsilon)) s << " (normalization problem ?)";
        s << ".";
        comments.push_back(s.str());
      }
    }
 
    return comments;
  }
 
  template < typename GUM_SCALAR >
  Tensor< GUM_SCALAR > IBayesNet< GUM_SCALAR >::evEq(const std::string& name, double value) const {
    return Tensor< GUM_SCALAR >::evEq(variableFromName(name), value);
  }
 
  template < typename GUM_SCALAR >
  Tensor< GUM_SCALAR >
      IBayesNet< GUM_SCALAR >::evIn(const std::string& name, double val1, double val2) const {
    return Tensor< GUM_SCALAR >::evIn(variableFromName(name), val1, val2);
  }
 
  template < typename GUM_SCALAR >
  Tensor< GUM_SCALAR > IBayesNet< GUM_SCALAR >::evGt(const std::string& name, double val) const {
    return Tensor< GUM_SCALAR >::evGt(variableFromName(name), val);
  }
 
  template < typename GUM_SCALAR >
  Tensor< GUM_SCALAR > IBayesNet< GUM_SCALAR >::evLt(const std::string& name, double val) const {
    return Tensor< GUM_SCALAR >::evLt(variableFromName(name), val);
  }
} /* namespace gum */