BayesNetFragment_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 <agrum/base/multidim/tensor.h>
#include <agrum/BN/BayesNet.h>
#include <agrum/BN/BayesNetFragment.h>
 
namespace gum {
  template < typename GUM_SCALAR >
  BayesNetFragment< GUM_SCALAR >::BayesNetFragment(const IBayesNet< GUM_SCALAR >& bn) :
      DiGraphListener(&bn.dag()), _bn_(bn) {
    GUM_CONSTRUCTOR(BayesNetFragment)
  }
 
  template < typename GUM_SCALAR >
  BayesNetFragment< GUM_SCALAR >::~BayesNetFragment() {
    GUM_DESTRUCTOR(BayesNetFragment)
 
    for (auto node: nodes())
      if (_localCPTs_.exists(node)) uninstallCPT_(node);
  }
 
  //============================================================
  // signals to keep consistency with the referred BayesNet
  template < typename GUM_SCALAR >
  INLINE void BayesNetFragment< GUM_SCALAR >::whenNodeAdded(const void* src, NodeId id) {
    // nothing to do
  }
 
  template < typename GUM_SCALAR >
  INLINE void BayesNetFragment< GUM_SCALAR >::whenNodeDeleted(const void* src, NodeId id) {
    uninstallNode(id);
  }
 
  template < typename GUM_SCALAR >
  INLINE void
      BayesNetFragment< GUM_SCALAR >::whenArcAdded(const void* src, NodeId from, NodeId to) {
    // nothing to do
  }
 
  template < typename GUM_SCALAR >
  INLINE void
      BayesNetFragment< GUM_SCALAR >::whenArcDeleted(const void* src, NodeId from, NodeId to) {
    if (dag().existsArc(from, to)) uninstallArc_(from, to);
  }
 
  //============================================================
  // IBayesNet interface : BayesNetFragment here is a decorator for the bn
 
  template < typename GUM_SCALAR >
  INLINE const Tensor< GUM_SCALAR >& BayesNetFragment< GUM_SCALAR >::cpt(NodeId id) const {
    if (!isInstalledNode(id)) GUM_ERROR(NotFound, "NodeId " << id << " is not installed")
 
    if (_localCPTs_.exists(id)) return *_localCPTs_[id];
    else return _bn_.cpt(id);
  }
 
  template < typename GUM_SCALAR >
  INLINE const VariableNodeMap& BayesNetFragment< GUM_SCALAR >::variableNodeMap() const {
    return this->_bn_.variableNodeMap();
  }
 
  template < typename GUM_SCALAR >
  INLINE const DiscreteVariable& BayesNetFragment< GUM_SCALAR >::variable(NodeId id) const {
    if (!isInstalledNode(id)) GUM_ERROR(NotFound, "NodeId " << id << " is not installed")
 
    return _bn_.variable(id);
  }
 
  template < typename GUM_SCALAR >
  INLINE NodeId BayesNetFragment< GUM_SCALAR >::nodeId(const DiscreteVariable& var) const {
    NodeId id = _bn_.nodeId(var);
 
    if (!isInstalledNode(id)) GUM_ERROR(NotFound, "variable " << var.name() << " is not installed")
 
    return id;
  }
 
  template < typename GUM_SCALAR >
  INLINE NodeId BayesNetFragment< GUM_SCALAR >::idFromName(const std::string& name) const {
    NodeId id = _bn_.idFromName(name);
 
    if (!isInstalledNode(id)) GUM_ERROR(NotFound, "variable " << name << " is not installed")
 
    return id;
  }
 
  template < typename GUM_SCALAR >
  INLINE const DiscreteVariable&
      BayesNetFragment< GUM_SCALAR >::variableFromName(const std::string& name) const {
    NodeId id = idFromName(name);
 
    if (!isInstalledNode(id)) GUM_ERROR(NotFound, "variable " << name << " is not installed")
 
    return _bn_.variable(id);
  }
 
  //============================================================
  // specific API for BayesNetFragment
  template < typename GUM_SCALAR >
  INLINE bool BayesNetFragment< GUM_SCALAR >::isInstalledNode(NodeId id) const {
    return dag().existsNode(id);
  }
 
  template < typename GUM_SCALAR >
  void BayesNetFragment< GUM_SCALAR >::installNode(NodeId id) {
    if (!_bn_.dag().existsNode(id))
      GUM_ERROR(NotFound, "Node " << id << " does not exist in referred BayesNet")
 
    if (!isInstalledNode(id)) {
      this->dag_.addNodeWithId(id);
 
      // adding arcs with id as a tail
      for (auto pa: this->_bn_.parents(id)) {
        if (isInstalledNode(pa)) this->dag_.addArc(pa, id);
      }
 
      // adding arcs with id as a head
      for (auto son: this->_bn_.children(id))
        if (isInstalledNode(son)) this->dag_.addArc(id, son);
    }
  }
 
  template < typename GUM_SCALAR >
  INLINE void BayesNetFragment< GUM_SCALAR >::installAscendants(NodeId id) {
    installNode(id);
 
    // bn is a dag => this will have an end ...
    for (auto pa: this->_bn_.parents(id))
      installAscendants(pa);
  }
 
  template < typename GUM_SCALAR >
  INLINE void BayesNetFragment< GUM_SCALAR >::uninstallNode(NodeId id) {
    if (isInstalledNode(id)) {
      uninstallCPT(id);
      this->dag_.eraseNode(id);
    }
  }
 
  template < typename GUM_SCALAR >
  INLINE void BayesNetFragment< GUM_SCALAR >::uninstallArc_(NodeId from, NodeId to) {
    this->dag_.eraseArc(Arc(from, to));
  }
 
  template < typename GUM_SCALAR >
  INLINE void BayesNetFragment< GUM_SCALAR >::installArc_(NodeId from, NodeId to) {
    this->dag_.addArc(from, to);
  }
 
  template < typename GUM_SCALAR >
  void BayesNetFragment< GUM_SCALAR >::installCPT_(NodeId id, const Tensor< GUM_SCALAR >& pot) {
    // topology
    const auto& parents = this->parents(id);
    for (auto node_it = parents.beginSafe(); node_it != parents.endSafe();
         ++node_it)   // safe iterator needed here
      uninstallArc_(*node_it, id);
 
    for (Idx i = 1; i < pot.nbrDim(); i++) {
      NodeId parent = _bn_.idFromName(pot.variable(i).name());
 
      if (isInstalledNode(parent)) installArc_(parent, id);
    }
 
    // local cpt
    if (_localCPTs_.exists(id)) uninstallCPT_(id);
 
    _localCPTs_.insert(id, new gum::Tensor< GUM_SCALAR >(pot));
  }
 
  template < typename GUM_SCALAR >
  void BayesNetFragment< GUM_SCALAR >::installCPT(NodeId id, const Tensor< GUM_SCALAR >& pot) {
    if (!dag().existsNode(id))
      GUM_ERROR(NotFound, "Node " << id << " is not installed in the fragment")
 
    if (&(pot.variable(0)) != &(variable(id))) {
      GUM_ERROR(OperationNotAllowed,
                "The tensor is not a marginal for  _bn_.variable <" << variable(id).name() << ">")
    }
 
    const NodeSet& parents = _bn_.parents(id);
 
    for (Idx i = 1; i < pot.nbrDim(); i++) {
      if (!parents.contains(_bn_.idFromName(pot.variable(i).name())))
        GUM_ERROR(OperationNotAllowed,
                  "Variable <" << pot.variable(i).name() << "> is not in the parents of node "
                               << id)
    }
 
    installCPT_(id, pot);
  }
 
  template < typename GUM_SCALAR >
  INLINE void BayesNetFragment< GUM_SCALAR >::uninstallCPT_(NodeId id) {
    delete _localCPTs_[id];
    _localCPTs_.erase(id);
  }
 
  template < typename GUM_SCALAR >
  INLINE void BayesNetFragment< GUM_SCALAR >::uninstallCPT(NodeId id) {
    if (_localCPTs_.exists(id)) {
      uninstallCPT_(id);
 
      // re-create arcs from referred tensor
      const Tensor< GUM_SCALAR >& pot = cpt(id);
 
      for (Idx i = 1; i < pot.nbrDim(); i++) {
        NodeId parent = _bn_.idFromName(pot.variable(i).name());
 
        if (isInstalledNode(parent)) installArc_(parent, id);
      }
    }
  }
 
  template < typename GUM_SCALAR >
  void BayesNetFragment< GUM_SCALAR >::installMarginal(NodeId id, const Tensor< GUM_SCALAR >& pot) {
    if (!isInstalledNode(id)) {
      GUM_ERROR(NotFound, "The node " << id << " is not part of this fragment")
    }
 
    if (pot.nbrDim() > 1) {
      GUM_ERROR(OperationNotAllowed, "The tensor is not a marginal :" << pot)
    }
 
    if (&(pot.variable(0)) != &(_bn_.variable(id))) {
      GUM_ERROR(OperationNotAllowed,
                "The tensor is not a marginal for  _bn_.variable <" << _bn_.variable(id).name()
                                                                    << ">")
    }
 
    installCPT_(id, pot);
  }
 
  template < typename GUM_SCALAR >
  bool BayesNetFragment< GUM_SCALAR >::checkConsistency(NodeId id) const {
    if (!isInstalledNode(id))
      GUM_ERROR(NotFound, "The node " << id << " is not part of this fragment")
 
    const auto& cpt = this->cpt(id);
    NodeSet     cpt_parents;
 
    for (Idx i = 1; i < cpt.nbrDim(); i++) {
      cpt_parents.insert(_bn_.idFromName(cpt.variable(i).name()));
    }
 
    return (this->parents(id) == cpt_parents);
  }
 
  template < typename GUM_SCALAR >
  INLINE bool BayesNetFragment< GUM_SCALAR >::checkConsistency() const {
    for (auto node: nodes())
      if (!checkConsistency(node)) return false;
 
    return true;
  }
 
  template < typename GUM_SCALAR >
  std::string BayesNetFragment< GUM_SCALAR >::toDot() const {
    std::stringstream output;
    output << "digraph \"";
 
    std::string bn_name;
 
    static std::string inFragmentStyle    = "fillcolor=\"#ffffaa\","
                                            "color=\"#000000\","
                                            "fontcolor=\"#000000\"";
    static std::string styleWithLocalCPT  = "fillcolor=\"#ffddaa\","
                                            "color=\"#000000\","
                                            "fontcolor=\"#000000\"";
    static std::string notConsistantStyle = "fillcolor=\"#ff0000\","
                                            "color=\"#000000\","
                                            "fontcolor=\"#ffff00\"";
    static std::string outFragmentStyle   = "fillcolor=\"#f0f0f0\","
                                            "color=\"#f0f0f0\","
                                            "fontcolor=\"#000000\"";
 
    try {
      bn_name = _bn_.property("name");
    } catch (NotFound const&) { bn_name = "no_name"; }
 
    bn_name = "Fragment of " + bn_name;
 
    output << bn_name << "\" {" << std::endl;
    output << "  graph [bgcolor=transparent,label=\"" << bn_name << "\"];" << std::endl;
    output << "  node [style=filled];" << std::endl << std::endl;
 
    for (auto node: _bn_.nodes()) {
      output << "\"" << _bn_.variable(node).name() << "\" [comment=\"" << node << ":"
             << _bn_.variable(node) << ", \"";
 
      if (isInstalledNode(node)) {
        if (!checkConsistency(node)) {
          output << notConsistantStyle;
        } else if (_localCPTs_.exists(node)) output << styleWithLocalCPT;
        else output << inFragmentStyle;
      } else output << outFragmentStyle;
 
      output << "];" << std::endl;
    }
 
    output << std::endl;
 
    std::string tab = "  ";
 
    for (auto node: _bn_.nodes()) {
      if (_bn_.children(node).size() > 0) {
        for (auto child: _bn_.children(node)) {
          output << tab << "\"" << _bn_.variable(node).name() << "\" -> "
                 << "\"" << _bn_.variable(child).name() << "\" [";
 
          if (dag().existsArc(Arc(node, child))) output << inFragmentStyle;
          else output << outFragmentStyle;
 
          output << "];" << std::endl;
        }
      }
    }
 
    output << "}" << std::endl;
 
    return output.str();
  }
 
  template < typename GUM_SCALAR >
  gum::BayesNet< GUM_SCALAR > BayesNetFragment< GUM_SCALAR >::toBN() const {
    if (!checkConsistency()) {
      GUM_ERROR(OperationNotAllowed, "The fragment contains un-consistent node(s)")
    }
    gum::BayesNet< GUM_SCALAR > res;
    for (const auto nod: nodes()) {
      res.add(variable(nod), nod);
    }
    for (const auto& arc: dag().arcs()) {
      res.addArc(arc.tail(), arc.head());
    }
    for (const auto nod: nodes()) {
      res.cpt(nod).fillWith(cpt(nod));
    }
 
    return res;
  }
}   // namespace gum