lemon/hao_orlin.h
author athos
Mon, 02 Oct 2006 11:18:30 +0000
changeset 2226 0411ac8a2d87
parent 2211 c790d04e192a
child 2228 f71b0f9a7c3a
permissions -rw-r--r--
MIP interface tested (and corrected) for cplex 9.0
     1 /* -*- C++ -*-
     2  *
     3  * This file is a part of LEMON, a generic C++ optimization library
     4  *
     5  * Copyright (C) 2003-2006
     6  * Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
     7  * (Egervary Research Group on Combinatorial Optimization, EGRES).
     8  *
     9  * Permission to use, modify and distribute this software is granted
    10  * provided that this copyright notice appears in all copies. For
    11  * precise terms see the accompanying LICENSE file.
    12  *
    13  * This software is provided "AS IS" with no warranty of any kind,
    14  * express or implied, and with no claim as to its suitability for any
    15  * purpose.
    16  *
    17  */
    18 
    19 #ifndef LEMON_HAO_ORLIN_H
    20 #define LEMON_HAO_ORLIN_H
    21 
    22 #include <vector>
    23 #include <queue>
    24 #include <limits>
    25 
    26 #include <lemon/maps.h>
    27 #include <lemon/graph_utils.h>
    28 #include <lemon/graph_adaptor.h>
    29 #include <lemon/iterable_maps.h>
    30  
    31 
    32 /// \file
    33 /// \ingroup flowalgs
    34 /// \brief Implementation of the Hao-Orlin algorithm.
    35 ///
    36 /// Implementation of the HaoOrlin algorithms class for testing network 
    37 /// reliability.
    38 
    39 namespace lemon {
    40 
    41   /// \ingroup flowalgs
    42   ///
    43   /// \brief %Hao-Orlin algorithm for calculate minimum cut in directed graphs.
    44   ///
    45   /// Hao-Orlin calculates the minimum cut in directed graphs. It
    46   /// separates the nodes of the graph into two disjoint sets, 
    47   /// \f$ V_{out} \f$ and \f$ V_{in} \f$. This separation is the minimum
    48   /// cut if the summary of the edge capacities which source is in
    49   /// \f$ V_{out} \f$ and the target is in \f$ V_{in} \f$ is the
    50   /// minimum.  The algorithm is a modified push-relabel preflow
    51   /// algorithm and it calculates the minimum cut in \f$ O(n^3) \f$
    52   /// time. The purpose of such algorithm is testing network
    53   /// reliability. For sparse undirected graph you can use the
    54   /// algorithm of Nagamochi and Ibaraki which solves the undirected
    55   /// problem in \f$ O(ne + n^2 \log(n)) \f$ time and it is implemented in the
    56   /// MinCut algorithm class.
    57   ///
    58   /// \param _Graph is the graph type of the algorithm.
    59   /// \param _CapacityMap is an edge map of capacities which should
    60   /// be any numreric type. The default type is _Graph::EdgeMap<int>.
    61   /// \param _Tolerance is the handler of the inexact computation. The
    62   /// default type for it is Tolerance<typename CapacityMap::Value>.
    63   ///
    64   /// \author Attila Bernath and Balazs Dezso
    65 #ifdef DOXYGEN
    66   template <typename _Graph, typename _CapacityMap, typename _Tolerance>
    67 #else
    68   template <typename _Graph,
    69 	    typename _CapacityMap = typename _Graph::template EdgeMap<int>,
    70             typename _Tolerance = Tolerance<typename _CapacityMap::Value> >
    71 #endif
    72   class HaoOrlin {
    73   protected:
    74 
    75     typedef _Graph Graph;
    76     typedef _CapacityMap CapacityMap;
    77     typedef _Tolerance Tolerance;
    78 
    79     typedef typename CapacityMap::Value Value;
    80 
    81     
    82     typedef typename Graph::Node Node;
    83     typedef typename Graph::NodeIt NodeIt;
    84     typedef typename Graph::EdgeIt EdgeIt;
    85     typedef typename Graph::OutEdgeIt OutEdgeIt;
    86     typedef typename Graph::InEdgeIt InEdgeIt;
    87 
    88     const Graph* _graph;
    89 
    90     const CapacityMap* _capacity;
    91 
    92     typedef typename Graph::template EdgeMap<Value> FlowMap;
    93 
    94     FlowMap* _preflow;
    95 
    96     Node _source, _target;
    97     int _node_num;
    98 
    99     typedef ResGraphAdaptor<const Graph, Value, CapacityMap, 
   100                             FlowMap, Tolerance> OutResGraph;
   101     typedef typename OutResGraph::Edge OutResEdge;
   102     
   103     OutResGraph* _out_res_graph;
   104 
   105     typedef typename Graph::template NodeMap<OutResEdge> OutCurrentEdgeMap;
   106     OutCurrentEdgeMap* _out_current_edge;  
   107 
   108     typedef RevGraphAdaptor<const Graph> RevGraph;
   109     RevGraph* _rev_graph;
   110 
   111     typedef ResGraphAdaptor<const RevGraph, Value, CapacityMap, 
   112                             FlowMap, Tolerance> InResGraph;
   113     typedef typename InResGraph::Edge InResEdge;
   114     
   115     InResGraph* _in_res_graph;
   116 
   117     typedef typename Graph::template NodeMap<InResEdge> InCurrentEdgeMap;
   118     InCurrentEdgeMap* _in_current_edge;  
   119 
   120 
   121     typedef IterableBoolMap<Graph, Node> WakeMap;
   122     WakeMap* _wake;
   123 
   124     typedef typename Graph::template NodeMap<int> DistMap;
   125     DistMap* _dist;  
   126     
   127     typedef typename Graph::template NodeMap<Value> ExcessMap;
   128     ExcessMap* _excess;
   129 
   130     typedef typename Graph::template NodeMap<bool> SourceSetMap;
   131     SourceSetMap* _source_set;
   132 
   133     std::vector<int> _level_size;
   134 
   135     int _highest_active;
   136     std::vector<std::list<Node> > _active_nodes;
   137 
   138     int _dormant_max;
   139     std::vector<std::list<Node> > _dormant;
   140 
   141 
   142     Value _min_cut;
   143 
   144     typedef typename Graph::template NodeMap<bool> MinCutMap;
   145     MinCutMap* _min_cut_map;
   146 
   147     Tolerance _tolerance;
   148 
   149   public: 
   150 
   151     /// \brief Constructor
   152     ///
   153     /// Constructor of the algorithm class. 
   154     HaoOrlin(const Graph& graph, const CapacityMap& capacity, 
   155              const Tolerance& tolerance = Tolerance()) :
   156       _graph(&graph), _capacity(&capacity), 
   157       _preflow(0), _source(), _target(), 
   158       _out_res_graph(0), _out_current_edge(0),
   159       _rev_graph(0), _in_res_graph(0), _in_current_edge(0),
   160       _wake(0),_dist(0), _excess(0), _source_set(0), 
   161       _highest_active(), _active_nodes(), _dormant_max(), _dormant(), 
   162       _min_cut(), _min_cut_map(0), _tolerance(tolerance) {}
   163 
   164     ~HaoOrlin() {
   165       if (_min_cut_map) {
   166         delete _min_cut_map;
   167       } 
   168       if (_in_current_edge) {
   169         delete _in_current_edge;
   170       }
   171       if (_in_res_graph) {
   172         delete _in_res_graph;
   173       }
   174       if (_rev_graph) {
   175         delete _rev_graph;
   176       }
   177       if (_out_current_edge) {
   178         delete _out_current_edge;
   179       }
   180       if (_out_res_graph) {
   181         delete _out_res_graph;
   182       }
   183       if (_source_set) {
   184         delete _source_set;
   185       }
   186       if (_excess) {
   187         delete _excess;
   188       }
   189       if (_dist) {
   190         delete _dist;
   191       }
   192       if (_wake) {
   193         delete _wake;
   194       }
   195       if (_preflow) {
   196         delete _preflow;
   197       }
   198     }
   199     
   200   private:
   201     
   202     template <typename ResGraph, typename EdgeMap>
   203     void findMinCut(const Node& target, bool out, 
   204                     ResGraph& res_graph, EdgeMap& current_edge) {
   205       typedef typename ResGraph::Edge ResEdge;
   206       typedef typename ResGraph::OutEdgeIt ResOutEdgeIt;
   207 
   208       for (typename Graph::EdgeIt it(*_graph); it != INVALID; ++it) {
   209         (*_preflow)[it] = 0;      
   210       }
   211       for (NodeIt it(*_graph); it != INVALID; ++it) {
   212         (*_wake)[it] = true;
   213         (*_dist)[it] = 1;
   214         (*_excess)[it] = 0;
   215         (*_source_set)[it] = false;
   216 
   217         res_graph.firstOut(current_edge[it], it);
   218       }
   219 
   220       _target = target;
   221       (*_dist)[target] = 0;
   222 
   223       for (ResOutEdgeIt it(res_graph, _source); it != INVALID; ++it) {
   224         Value delta = res_graph.rescap(it);
   225         if (!_tolerance.positive(delta)) continue;
   226         
   227         (*_excess)[res_graph.source(it)] -= delta;
   228         res_graph.augment(it, delta);
   229         Node a = res_graph.target(it);
   230         if (!_tolerance.positive((*_excess)[a]) && 
   231             (*_wake)[a] && a != _target) {
   232           _active_nodes[(*_dist)[a]].push_front(a);
   233           if (_highest_active < (*_dist)[a]) {
   234             _highest_active = (*_dist)[a];
   235           }
   236         }
   237         (*_excess)[a] += delta;
   238       }
   239 
   240       _dormant[0].push_front(_source);
   241       (*_source_set)[_source] = true;
   242       _dormant_max = 0;
   243       (*_wake)[_source] = false;
   244 
   245       _level_size[0] = 1;
   246       _level_size[1] = _node_num - 1;
   247 
   248       do {
   249 	Node n;
   250 	while ((n = findActiveNode()) != INVALID) {
   251 	  ResEdge e;
   252 	  while (_tolerance.positive((*_excess)[n]) && 
   253                  (e = findAdmissibleEdge(n, res_graph, current_edge)) 
   254                  != INVALID){
   255 	    Value delta;
   256 	    if ((*_excess)[n] < res_graph.rescap(e)) {
   257 	      delta = (*_excess)[n];
   258 	    } else {
   259 	      delta = res_graph.rescap(e);
   260 	      res_graph.nextOut(current_edge[n]);
   261 	    }
   262             if (!_tolerance.positive(delta)) continue;
   263 	    res_graph.augment(e, delta);
   264 	    (*_excess)[res_graph.source(e)] -= delta;
   265 	    Node a = res_graph.target(e);
   266 	    if (!_tolerance.positive((*_excess)[a]) && a != _target) {
   267 	      _active_nodes[(*_dist)[a]].push_front(a);
   268 	    }
   269 	    (*_excess)[a] += delta;
   270 	  }
   271 	  if (_tolerance.positive((*_excess)[n])) {
   272 	    relabel(n, res_graph, current_edge);
   273           }
   274 	}
   275 
   276 	Value current_value = cutValue(out);
   277  	if (_min_cut > current_value){
   278           if (out) {
   279             for (NodeIt it(*_graph); it != INVALID; ++it) {
   280               _min_cut_map->set(it, !(*_wake)[it]);
   281             } 
   282           } else {
   283             for (NodeIt it(*_graph); it != INVALID; ++it) {
   284               _min_cut_map->set(it, (*_wake)[it]);
   285             } 
   286           }
   287 
   288 	  _min_cut = current_value;
   289  	}
   290 
   291       } while (selectNewSink(res_graph));
   292     }
   293 
   294     template <typename ResGraph, typename EdgeMap>
   295     void relabel(const Node& n, ResGraph& res_graph, EdgeMap& current_edge) {
   296       typedef typename ResGraph::OutEdgeIt ResOutEdgeIt;
   297 
   298       int k = (*_dist)[n];
   299       if (_level_size[k] == 1) {
   300 	++_dormant_max;
   301 	for (NodeIt it(*_graph); it != INVALID; ++it) {
   302 	  if ((*_wake)[it] && (*_dist)[it] >= k) {
   303 	    (*_wake)[it] = false;
   304 	    _dormant[_dormant_max].push_front(it);
   305 	    --_level_size[(*_dist)[it]];
   306 	  }
   307 	}
   308 	--_highest_active;
   309       } else {	
   310         int new_dist = _node_num;
   311         for (ResOutEdgeIt e(res_graph, n); e != INVALID; ++e) {
   312           Node t = res_graph.target(e);
   313           if ((*_wake)[t] && new_dist > (*_dist)[t]) {
   314             new_dist = (*_dist)[t];
   315           }
   316         }
   317         if (new_dist == _node_num) {
   318 	  ++_dormant_max;
   319 	  (*_wake)[n] = false;
   320 	  _dormant[_dormant_max].push_front(n);
   321 	  --_level_size[(*_dist)[n]];
   322 	} else {	    
   323 	  --_level_size[(*_dist)[n]];
   324 	  (*_dist)[n] = new_dist + 1;
   325 	  _highest_active = (*_dist)[n];
   326 	  _active_nodes[_highest_active].push_front(n);
   327 	  ++_level_size[(*_dist)[n]];
   328 	  res_graph.firstOut(current_edge[n], n);
   329 	}
   330       }
   331     }
   332 
   333     template <typename ResGraph>
   334     bool selectNewSink(ResGraph& res_graph) {
   335       typedef typename ResGraph::OutEdgeIt ResOutEdgeIt;
   336 
   337       Node old_target = _target;
   338       (*_wake)[_target] = false;
   339       --_level_size[(*_dist)[_target]];
   340       _dormant[0].push_front(_target);
   341       (*_source_set)[_target] = true;
   342       if ((int)_dormant[0].size() == _node_num){
   343         _dormant[0].clear();
   344 	return false;
   345       }
   346 
   347       bool wake_was_empty = false;
   348 
   349       if(_wake->trueNum() == 0) {
   350 	while (!_dormant[_dormant_max].empty()){
   351 	  (*_wake)[_dormant[_dormant_max].front()] = true;
   352 	  ++_level_size[(*_dist)[_dormant[_dormant_max].front()]];
   353 	  _dormant[_dormant_max].pop_front();
   354 	}
   355 	--_dormant_max;
   356 	wake_was_empty = true;
   357       }
   358 
   359       int min_dist = std::numeric_limits<int>::max();
   360       for (typename WakeMap::TrueIt it(*_wake); it != INVALID; ++it) {
   361 	if (min_dist > (*_dist)[it]){
   362 	  _target = it;
   363 	  min_dist = (*_dist)[it];
   364 	}
   365       }
   366 
   367       if (wake_was_empty){
   368 	for (typename WakeMap::TrueIt it(*_wake); it != INVALID; ++it) {
   369           if (_tolerance.positive((*_excess)[it])) {
   370 	    if ((*_wake)[it] && it != _target) {
   371 	      _active_nodes[(*_dist)[it]].push_front(it);
   372             }
   373 	    if (_highest_active < (*_dist)[it]) {
   374 	      _highest_active = (*_dist)[it];		    
   375             }
   376 	  }
   377 	}
   378       }
   379 
   380       for (ResOutEdgeIt e(res_graph, old_target); e!=INVALID; ++e){
   381 	if (!(*_source_set)[res_graph.target(e)]) {
   382           Value delta = res_graph.rescap(e);
   383           if (!_tolerance.positive(delta)) continue;
   384           res_graph.augment(e, delta);
   385           (*_excess)[res_graph.source(e)] -= delta;
   386           Node a = res_graph.target(e);
   387           if (!_tolerance.positive((*_excess)[a]) && 
   388               (*_wake)[a] && a != _target) {
   389             _active_nodes[(*_dist)[a]].push_front(a);
   390             if (_highest_active < (*_dist)[a]) {
   391               _highest_active = (*_dist)[a];
   392             }
   393           }
   394           (*_excess)[a] += delta;
   395 	}
   396       }
   397       
   398       return true;
   399     }
   400     
   401     Node findActiveNode() {
   402       while (_highest_active > 0 && _active_nodes[_highest_active].empty()){ 
   403 	--_highest_active;
   404       }
   405       if( _highest_active > 0) {
   406        	Node n = _active_nodes[_highest_active].front();
   407 	_active_nodes[_highest_active].pop_front();
   408 	return n;
   409       } else {
   410 	return INVALID;
   411       }
   412     }
   413 
   414     template <typename ResGraph, typename EdgeMap>
   415     typename ResGraph::Edge findAdmissibleEdge(const Node& n, 
   416                                                ResGraph& res_graph, 
   417                                                EdgeMap& current_edge) {
   418       typedef typename ResGraph::Edge ResEdge;
   419       ResEdge e = current_edge[n];
   420       while (e != INVALID && 
   421              ((*_dist)[n] <= (*_dist)[res_graph.target(e)] || 
   422               !(*_wake)[res_graph.target(e)])) {
   423 	res_graph.nextOut(e);
   424       }
   425       if (e != INVALID) {
   426 	current_edge[n] = e;	
   427 	return e;
   428       } else {
   429 	return INVALID;
   430       }
   431     }
   432 
   433     Value cutValue(bool out) {
   434       Value value = 0;
   435       if (out) {
   436         for (typename WakeMap::TrueIt it(*_wake); it != INVALID; ++it) {
   437           for (InEdgeIt e(*_graph, it); e != INVALID; ++e) {
   438             if (!(*_wake)[_graph->source(e)]){
   439               value += (*_capacity)[e];
   440             }	
   441           }
   442         }
   443       } else {
   444         for (typename WakeMap::TrueIt it(*_wake); it != INVALID; ++it) {
   445           for (OutEdgeIt e(*_graph, it); e != INVALID; ++e) {
   446             if (!(*_wake)[_graph->target(e)]){
   447               value += (*_capacity)[e];
   448             }	
   449           }
   450         }
   451       }
   452       return value;
   453     }
   454 
   455 
   456   public:
   457 
   458     /// \name Execution control
   459     /// The simplest way to execute the algorithm is to use
   460     /// one of the member functions called \c run(...).
   461     /// \n
   462     /// If you need more control on the execution,
   463     /// first you must call \ref init(), then the \ref calculateIn() or
   464     /// \ref calculateIn() functions.
   465 
   466     /// @{
   467 
   468     /// \brief Initializes the internal data structures.
   469     ///
   470     /// Initializes the internal data structures. It creates
   471     /// the maps, residual graph adaptors and some bucket structures
   472     /// for the algorithm. 
   473     void init() {
   474       init(NodeIt(*_graph));
   475     }
   476 
   477     /// \brief Initializes the internal data structures.
   478     ///
   479     /// Initializes the internal data structures. It creates
   480     /// the maps, residual graph adaptor and some bucket structures
   481     /// for the algorithm. The \c source node is used as the push-relabel
   482     /// algorithm's source.
   483     void init(const Node& source) {
   484       _source = source;
   485       _node_num = countNodes(*_graph);
   486 
   487       _dormant.resize(_node_num);
   488       _level_size.resize(_node_num, 0);
   489       _active_nodes.resize(_node_num);
   490 
   491       if (!_preflow) {
   492         _preflow = new FlowMap(*_graph);
   493       }
   494       if (!_wake) {
   495         _wake = new WakeMap(*_graph);
   496       }
   497       if (!_dist) {
   498         _dist = new DistMap(*_graph);
   499       }
   500       if (!_excess) {
   501         _excess = new ExcessMap(*_graph);
   502       }
   503       if (!_source_set) {
   504         _source_set = new SourceSetMap(*_graph);
   505       }
   506       if (!_out_res_graph) {
   507         _out_res_graph = new OutResGraph(*_graph, *_capacity, *_preflow);
   508       }
   509       if (!_out_current_edge) {
   510         _out_current_edge = new OutCurrentEdgeMap(*_graph);
   511       }
   512       if (!_rev_graph) {
   513         _rev_graph = new RevGraph(*_graph);
   514       }
   515       if (!_in_res_graph) {
   516         _in_res_graph = new InResGraph(*_rev_graph, *_capacity, *_preflow);
   517       }
   518       if (!_in_current_edge) {
   519         _in_current_edge = new InCurrentEdgeMap(*_graph);
   520       }
   521       if (!_min_cut_map) {
   522         _min_cut_map = new MinCutMap(*_graph);
   523       }
   524 
   525       _min_cut = std::numeric_limits<Value>::max();
   526     }
   527 
   528 
   529     /// \brief Calculates the minimum cut with the \c source node
   530     /// in the \f$ V_{out} \f$.
   531     ///
   532     /// Calculates the minimum cut with the \c source node
   533     /// in the \f$ V_{out} \f$.
   534     void calculateOut() {
   535       for (NodeIt it(*_graph); it != INVALID; ++it) {
   536         if (it != _source) {
   537           calculateOut(it);
   538           return;
   539         }
   540       }
   541     }
   542 
   543     /// \brief Calculates the minimum cut with the \c source node
   544     /// in the \f$ V_{out} \f$.
   545     ///
   546     /// Calculates the minimum cut with the \c source node
   547     /// in the \f$ V_{out} \f$. The \c target is the initial target
   548     /// for the push-relabel algorithm.
   549     void calculateOut(const Node& target) {
   550       findMinCut(target, true, *_out_res_graph, *_out_current_edge);
   551     }
   552 
   553     /// \brief Calculates the minimum cut with the \c source node
   554     /// in the \f$ V_{in} \f$.
   555     ///
   556     /// Calculates the minimum cut with the \c source node
   557     /// in the \f$ V_{in} \f$.
   558     void calculateIn() {
   559       for (NodeIt it(*_graph); it != INVALID; ++it) {
   560         if (it != _source) {
   561           calculateIn(it);
   562           return;
   563         }
   564       }
   565     }
   566 
   567     /// \brief Calculates the minimum cut with the \c source node
   568     /// in the \f$ V_{in} \f$.
   569     ///
   570     /// Calculates the minimum cut with the \c source node
   571     /// in the \f$ V_{in} \f$. The \c target is the initial target
   572     /// for the push-relabel algorithm.
   573     void calculateIn(const Node& target) {
   574       findMinCut(target, false, *_in_res_graph, *_in_current_edge);
   575     }
   576 
   577     /// \brief Runs the algorithm.
   578     ///
   579     /// Runs the algorithm. It finds a proper \c source and \c target
   580     /// and then calls the \ref init(), \ref calculateOut() and \ref
   581     /// calculateIn().
   582     void run() {
   583       init();
   584       for (NodeIt it(*_graph); it != INVALID; ++it) {
   585         if (it != _source) {
   586           calculateOut(it);
   587           calculateIn(it);
   588           return;
   589         }
   590       }
   591     }
   592 
   593     /// \brief Runs the algorithm.
   594     ///
   595     /// Runs the algorithm. It finds a proper \c target and then calls
   596     /// the \ref init(), \ref calculateOut() and \ref calculateIn().
   597     void run(const Node& s) {
   598       init(s);
   599       for (NodeIt it(*_graph); it != INVALID; ++it) {
   600         if (it != _source) {
   601           calculateOut(it);
   602           calculateIn(it);
   603           return;
   604         }
   605       }
   606     }
   607 
   608     /// \brief Runs the algorithm.
   609     ///
   610     /// Runs the algorithm. It just calls the \ref init() and then
   611     /// \ref calculateOut() and \ref calculateIn().
   612     void run(const Node& s, const Node& t) {
   613       init(s); 
   614       calculateOut(t);
   615       calculateIn(t);
   616     }
   617 
   618     /// @}
   619     
   620     /// \name Query Functions The result of the %HaoOrlin algorithm
   621     /// can be obtained using these functions.
   622     /// \n
   623     /// Before the use of these functions, either \ref run(), \ref
   624     /// calculateOut() or \ref calculateIn() must be called.
   625     
   626     /// @{
   627 
   628     /// \brief Returns the value of the minimum value cut.
   629     /// 
   630     /// Returns the value of the minimum value cut.
   631     Value minCut() const {
   632       return _min_cut;
   633     }
   634 
   635 
   636     /// \brief Returns a minimum value cut.
   637     ///
   638     /// Sets \c nodeMap to the characteristic vector of a minimum
   639     /// value cut. The nodes in \f$ V_{out} \f$ will be set true and
   640     /// the nodes in \f$ V_{in} \f$ will be set false. 
   641     /// \pre nodeMap should be a bool-valued node-map.
   642     template <typename NodeMap>
   643     Value minCut(NodeMap& nodeMap) const {
   644       for (NodeIt it(*_graph); it != INVALID; ++it) {
   645 	nodeMap.set(it, (*_min_cut_map)[it]);
   646       }
   647       return minCut();
   648     }
   649 
   650     /// @}
   651     
   652   }; //class HaoOrlin 
   653 
   654 
   655 } //namespace lemon
   656 
   657 #endif //LEMON_HAO_ORLIN_H