lemon/hao_orlin.h
author deba
Tue, 30 Oct 2007 20:21:10 +0000
changeset 2505 1bb471764ab8
parent 2386 81b47fc5c444
child 2530 f86f7e4eb2ba
permissions -rw-r--r--
Redesign interface of MaxMatching and UnionFindEnum
New class ExtendFindEnum

Faster MaxMatching
     1 /* -*- C++ -*-
     2  *
     3  * This file is a part of LEMON, a generic C++ optimization library
     4  *
     5  * Copyright (C) 2003-2007
     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 <cassert>
    23  
    24 
    25 
    26 #include <vector>
    27 #include <queue>
    28 #include <list>
    29 #include <limits>
    30 
    31 #include <lemon/maps.h>
    32 #include <lemon/graph_utils.h>
    33 #include <lemon/graph_adaptor.h>
    34 #include <lemon/iterable_maps.h>
    35 
    36 /// \file
    37 /// \ingroup min_cut
    38 /// \brief Implementation of the Hao-Orlin algorithm.
    39 ///
    40 /// Implementation of the HaoOrlin algorithms class for testing network 
    41 /// reliability.
    42 
    43 namespace lemon {
    44 
    45   /// \ingroup min_cut
    46   ///
    47   /// \brief %Hao-Orlin algorithm to find a minimum cut in directed graphs.
    48   ///
    49   /// Hao-Orlin calculates a minimum cut in a directed graph 
    50   /// \f$ D=(V,A) \f$. It takes a fixed node \f$ source \in V \f$ and consists
    51   /// of two phases: in the first phase it determines a minimum cut
    52   /// with \f$ source \f$ on the source-side (i.e. a set \f$ X\subsetneq V \f$
    53   /// with \f$ source \in X \f$ and minimal out-degree) and in the
    54   /// second phase it determines a minimum cut with \f$ source \f$ on the
    55   /// sink-side (i.e. a set \f$ X\subsetneq V \f$ with \f$ source \notin X \f$
    56   /// and minimal out-degree). Obviously, the smaller of these two
    57   /// cuts will be a minimum cut of \f$ D \f$. The algorithm is a
    58   /// modified push-relabel preflow algorithm and our implementation
    59   /// calculates the minimum cut in \f$ O(n^3) \f$ time (we use the
    60   /// highest-label rule). The purpose of such an algorithm is testing
    61   /// network reliability. For an undirected graph with \f$ n \f$
    62   /// nodes and \f$ e \f$ edges you can use the algorithm of Nagamochi
    63   /// and Ibaraki which solves the undirected problem in 
    64   /// \f$ O(ne + n^2 \log(n)) \f$ time: it is implemented in the MinCut 
    65   /// algorithm
    66   /// class.
    67   ///
    68   /// \param _Graph is the graph type of the algorithm.
    69   /// \param _CapacityMap is an edge map of capacities which should
    70   /// be any numreric type. The default type is _Graph::EdgeMap<int>.
    71   /// \param _Tolerance is the handler of the inexact computation. The
    72   /// default type for this is Tolerance<typename CapacityMap::Value>.
    73   ///
    74   /// \author Attila Bernath and Balazs Dezso
    75 #ifdef DOXYGEN
    76   template <typename _Graph, typename _CapacityMap, typename _Tolerance>
    77 #else
    78   template <typename _Graph,
    79 	    typename _CapacityMap = typename _Graph::template EdgeMap<int>,
    80             typename _Tolerance = Tolerance<typename _CapacityMap::Value> >
    81 #endif
    82   class HaoOrlin {
    83   protected:
    84 
    85     typedef _Graph Graph;
    86     typedef _CapacityMap CapacityMap;
    87     typedef _Tolerance Tolerance;
    88 
    89     typedef typename CapacityMap::Value Value;
    90 
    91     
    92     typedef typename Graph::Node Node;
    93     typedef typename Graph::NodeIt NodeIt;
    94     typedef typename Graph::EdgeIt EdgeIt;
    95     typedef typename Graph::OutEdgeIt OutEdgeIt;
    96     typedef typename Graph::InEdgeIt InEdgeIt;
    97 
    98     const Graph* _graph;
    99 
   100     const CapacityMap* _capacity;
   101 
   102     typedef typename Graph::template EdgeMap<Value> FlowMap;
   103 
   104     FlowMap* _preflow;
   105 
   106     Node _source, _target;
   107     int _node_num;
   108 
   109     typedef ResGraphAdaptor<const Graph, Value, CapacityMap, 
   110                             FlowMap, Tolerance> OutResGraph;
   111     typedef typename OutResGraph::Edge OutResEdge;
   112     
   113     OutResGraph* _out_res_graph;
   114 
   115     typedef RevGraphAdaptor<const Graph> RevGraph;
   116     RevGraph* _rev_graph;
   117 
   118     typedef ResGraphAdaptor<const RevGraph, Value, CapacityMap, 
   119                             FlowMap, Tolerance> InResGraph;
   120     typedef typename InResGraph::Edge InResEdge;
   121     
   122     InResGraph* _in_res_graph;
   123 
   124     typedef IterableBoolMap<Graph, Node> WakeMap;
   125     WakeMap* _wake;
   126 
   127     typedef typename Graph::template NodeMap<int> DistMap;
   128     DistMap* _dist;  
   129     
   130     typedef typename Graph::template NodeMap<Value> ExcessMap;
   131     ExcessMap* _excess;
   132 
   133     typedef typename Graph::template NodeMap<bool> SourceSetMap;
   134     SourceSetMap* _source_set;
   135 
   136     std::vector<int> _level_size;
   137 
   138     int _highest_active;
   139     std::vector<std::list<Node> > _active_nodes;
   140 
   141     int _dormant_max;
   142     std::vector<std::list<Node> > _dormant;
   143 
   144 
   145     Value _min_cut;
   146 
   147     typedef typename Graph::template NodeMap<bool> MinCutMap;
   148     MinCutMap* _min_cut_map;
   149 
   150     Tolerance _tolerance;
   151 
   152   public: 
   153 
   154     /// \brief Constructor
   155     ///
   156     /// Constructor of the algorithm class. 
   157     HaoOrlin(const Graph& graph, const CapacityMap& capacity, 
   158              const Tolerance& tolerance = Tolerance()) :
   159       _graph(&graph), _capacity(&capacity), 
   160       _preflow(0), _source(), _target(), 
   161       _out_res_graph(0), _rev_graph(0), _in_res_graph(0),
   162       _wake(0),_dist(0), _excess(0), _source_set(0), 
   163       _highest_active(), _active_nodes(), _dormant_max(), _dormant(), 
   164       _min_cut(), _min_cut_map(0), _tolerance(tolerance) {}
   165 
   166     ~HaoOrlin() {
   167       if (_min_cut_map) {
   168         delete _min_cut_map;
   169       } 
   170       if (_in_res_graph) {
   171         delete _in_res_graph;
   172       }
   173       if (_rev_graph) {
   174         delete _rev_graph;
   175       }
   176       if (_out_res_graph) {
   177         delete _out_res_graph;
   178       }
   179       if (_source_set) {
   180         delete _source_set;
   181       }
   182       if (_excess) {
   183         delete _excess;
   184       }
   185       if (_dist) {
   186         delete _dist;
   187       }
   188       if (_wake) {
   189         delete _wake;
   190       }
   191       if (_preflow) {
   192         delete _preflow;
   193       }
   194     }
   195     
   196   private:
   197     
   198     template <typename ResGraph>
   199     void findMinCut(const Node& target, bool out, ResGraph& res_graph) {
   200       typedef typename ResGraph::Edge ResEdge;
   201       typedef typename ResGraph::OutEdgeIt ResOutEdgeIt;
   202 
   203       for (typename Graph::EdgeIt it(*_graph); it != INVALID; ++it) {
   204         (*_preflow)[it] = 0;      
   205       }
   206       for (NodeIt it(*_graph); it != INVALID; ++it) {
   207         (*_wake)[it] = true;
   208         (*_dist)[it] = 1;
   209         (*_excess)[it] = 0;
   210         (*_source_set)[it] = false;
   211       }
   212 
   213       _dormant[0].push_front(_source);
   214       (*_source_set)[_source] = true;
   215       _dormant_max = 0;
   216       (*_wake)[_source] = false;
   217 
   218       _level_size[0] = 1;
   219       _level_size[1] = _node_num - 1;
   220 
   221       _target = target;
   222       (*_dist)[target] = 0;
   223 
   224       for (ResOutEdgeIt it(res_graph, _source); it != INVALID; ++it) {
   225         Value delta = res_graph.rescap(it);
   226         (*_excess)[_source] -= delta;
   227         res_graph.augment(it, delta);
   228         Node a = res_graph.target(it);
   229         if ((*_excess)[a] == 0 && (*_wake)[a] && a != _target) {
   230           _active_nodes[(*_dist)[a]].push_front(a);
   231           if (_highest_active < (*_dist)[a]) {
   232             _highest_active = (*_dist)[a];
   233           }
   234         }
   235         (*_excess)[a] += delta;
   236       }
   237 
   238 
   239       do {
   240 	Node n;
   241 	while ((n = findActiveNode()) != INVALID) {
   242 	  for (ResOutEdgeIt e(res_graph, n); e != INVALID; ++e) {
   243             Node a = res_graph.target(e);
   244             if ((*_dist)[a] >= (*_dist)[n] || !(*_wake)[a]) continue;
   245 	    Value delta = res_graph.rescap(e);
   246 	    if (_tolerance.positive((*_excess)[n] - delta)) {
   247               (*_excess)[n] -= delta;
   248 	    } else {
   249 	      delta = (*_excess)[n];
   250               (*_excess)[n] = 0;
   251 	    }
   252 	    res_graph.augment(e, delta);
   253 	    if ((*_excess)[a] == 0 && a != _target) {
   254 	      _active_nodes[(*_dist)[a]].push_front(a);
   255 	    }
   256 	    (*_excess)[a] += delta;
   257             if ((*_excess)[n] == 0) break;
   258 	  }
   259 	  if ((*_excess)[n] != 0) {
   260 	    relabel(n, res_graph);
   261           }
   262 	}
   263 
   264 	Value current_value = cutValue(out);
   265  	if (_min_cut > current_value){
   266           if (out) {
   267             for (NodeIt it(*_graph); it != INVALID; ++it) {
   268               _min_cut_map->set(it, !(*_wake)[it]);
   269             } 
   270           } else {
   271             for (NodeIt it(*_graph); it != INVALID; ++it) {
   272               _min_cut_map->set(it, (*_wake)[it]);
   273             } 
   274           }
   275 
   276 	  _min_cut = current_value;
   277  	}
   278 
   279       } while (selectNewSink(res_graph));
   280     }
   281 
   282     template <typename ResGraph>
   283     void relabel(const Node& n, ResGraph& res_graph) {
   284       typedef typename ResGraph::OutEdgeIt ResOutEdgeIt;
   285 
   286       int k = (*_dist)[n];
   287       if (_level_size[k] == 1) {
   288 	++_dormant_max;
   289 	for (NodeIt it(*_graph); it != INVALID; ++it) {
   290 	  if ((*_wake)[it] && (*_dist)[it] >= k) {
   291 	    (*_wake)[it] = false;
   292 	    _dormant[_dormant_max].push_front(it);
   293 	    --_level_size[(*_dist)[it]];
   294 	  }
   295 	}
   296         --_highest_active;
   297       } else {	
   298         int new_dist = _node_num;
   299         for (ResOutEdgeIt e(res_graph, n); e != INVALID; ++e) {
   300           Node t = res_graph.target(e);
   301           if ((*_wake)[t] && new_dist > (*_dist)[t]) {
   302             new_dist = (*_dist)[t];
   303           }
   304         }
   305         if (new_dist == _node_num) {
   306 	  ++_dormant_max;
   307 	  (*_wake)[n] = false;
   308 	  _dormant[_dormant_max].push_front(n);
   309 	  --_level_size[(*_dist)[n]];
   310 	} else {	    
   311 	  --_level_size[(*_dist)[n]];
   312 	  (*_dist)[n] = new_dist + 1;
   313 	  _highest_active = (*_dist)[n];
   314 	  _active_nodes[_highest_active].push_front(n);
   315 	  ++_level_size[(*_dist)[n]];
   316 	}
   317       }
   318     }
   319 
   320     template <typename ResGraph>
   321     bool selectNewSink(ResGraph& res_graph) {
   322       typedef typename ResGraph::OutEdgeIt ResOutEdgeIt;
   323 
   324       Node old_target = _target;
   325       (*_wake)[_target] = false;
   326       --_level_size[(*_dist)[_target]];
   327       _dormant[0].push_front(_target);
   328       (*_source_set)[_target] = true;
   329       if (int(_dormant[0].size()) == _node_num){
   330         _dormant[0].clear();
   331 	return false;
   332       }
   333 
   334       bool wake_was_empty = false;
   335 
   336       if(_wake->trueNum() == 0) {
   337 	while (!_dormant[_dormant_max].empty()){
   338 	  (*_wake)[_dormant[_dormant_max].front()] = true;
   339 	  ++_level_size[(*_dist)[_dormant[_dormant_max].front()]];
   340 	  _dormant[_dormant_max].pop_front();
   341 	}
   342 	--_dormant_max;
   343 	wake_was_empty = true;
   344       }
   345 
   346       int min_dist = std::numeric_limits<int>::max();
   347       for (typename WakeMap::TrueIt it(*_wake); it != INVALID; ++it) {
   348 	if (min_dist > (*_dist)[it]){
   349 	  _target = it;
   350 	  min_dist = (*_dist)[it];
   351 	}
   352       }
   353 
   354       if (wake_was_empty){
   355 	for (typename WakeMap::TrueIt it(*_wake); it != INVALID; ++it) {
   356           if ((*_excess)[it] != 0 && it != _target) {
   357             _active_nodes[(*_dist)[it]].push_front(it);
   358             if (_highest_active < (*_dist)[it]) {
   359               _highest_active = (*_dist)[it];		    
   360             }
   361 	  }
   362 	}
   363       }
   364 
   365       Node n = old_target;
   366       for (ResOutEdgeIt e(res_graph, n); e != INVALID; ++e){
   367         Node a = res_graph.target(e);
   368 	if (!(*_wake)[a]) continue;
   369         Value delta = res_graph.rescap(e);
   370         res_graph.augment(e, delta);
   371         (*_excess)[n] -= delta;
   372         if ((*_excess)[a] == 0 && (*_wake)[a] && a != _target) {
   373           _active_nodes[(*_dist)[a]].push_front(a);
   374           if (_highest_active < (*_dist)[a]) {
   375             _highest_active = (*_dist)[a];
   376           }
   377         }
   378         (*_excess)[a] += delta;
   379       }
   380       
   381       return true;
   382     }
   383 
   384     Node findActiveNode() {
   385       while (_highest_active > 0 && _active_nodes[_highest_active].empty()){ 
   386 	--_highest_active;
   387       }
   388       if( _highest_active > 0) {
   389        	Node n = _active_nodes[_highest_active].front();
   390 	_active_nodes[_highest_active].pop_front();
   391 	return n;
   392       } else {
   393 	return INVALID;
   394       }
   395     }
   396 
   397     Value cutValue(bool out) {
   398       Value value = 0;
   399       if (out) {
   400         for (typename WakeMap::TrueIt it(*_wake); it != INVALID; ++it) {
   401           for (InEdgeIt e(*_graph, it); e != INVALID; ++e) {
   402             if (!(*_wake)[_graph->source(e)]){
   403               value += (*_capacity)[e];
   404             }	
   405           }
   406         }
   407       } else {
   408         for (typename WakeMap::TrueIt it(*_wake); it != INVALID; ++it) {
   409           for (OutEdgeIt e(*_graph, it); e != INVALID; ++e) {
   410             if (!(*_wake)[_graph->target(e)]){
   411               value += (*_capacity)[e];
   412             }	
   413           }
   414         }
   415       }
   416       return value;
   417     }
   418 
   419 
   420   public:
   421 
   422     /// \name Execution control
   423     /// The simplest way to execute the algorithm is to use
   424     /// one of the member functions called \c run(...).
   425     /// \n
   426     /// If you need more control on the execution,
   427     /// first you must call \ref init(), then the \ref calculateIn() or
   428     /// \ref calculateIn() functions.
   429 
   430     /// @{
   431 
   432     /// \brief Initializes the internal data structures.
   433     ///
   434     /// Initializes the internal data structures. It creates
   435     /// the maps, residual graph adaptors and some bucket structures
   436     /// for the algorithm. 
   437     void init() {
   438       init(NodeIt(*_graph));
   439     }
   440 
   441     /// \brief Initializes the internal data structures.
   442     ///
   443     /// Initializes the internal data structures. It creates
   444     /// the maps, residual graph adaptor and some bucket structures
   445     /// for the algorithm. Node \c source  is used as the push-relabel
   446     /// algorithm's source.
   447     void init(const Node& source) {
   448       _source = source;
   449       _node_num = countNodes(*_graph);
   450 
   451       _dormant.resize(_node_num);
   452       _level_size.resize(_node_num, 0);
   453       _active_nodes.resize(_node_num);
   454 
   455       if (!_preflow) {
   456         _preflow = new FlowMap(*_graph);
   457       }
   458       if (!_wake) {
   459         _wake = new WakeMap(*_graph);
   460       }
   461       if (!_dist) {
   462         _dist = new DistMap(*_graph);
   463       }
   464       if (!_excess) {
   465         _excess = new ExcessMap(*_graph);
   466       }
   467       if (!_source_set) {
   468         _source_set = new SourceSetMap(*_graph);
   469       }
   470       if (!_out_res_graph) {
   471         _out_res_graph = new OutResGraph(*_graph, *_capacity, *_preflow);
   472       }
   473       if (!_rev_graph) {
   474         _rev_graph = new RevGraph(*_graph);
   475       }
   476       if (!_in_res_graph) {
   477         _in_res_graph = new InResGraph(*_rev_graph, *_capacity, *_preflow);
   478       }
   479       if (!_min_cut_map) {
   480         _min_cut_map = new MinCutMap(*_graph);
   481       }
   482 
   483       _min_cut = std::numeric_limits<Value>::max();
   484     }
   485 
   486 
   487     /// \brief Calculates a minimum cut with \f$ source \f$ on the
   488     /// source-side.
   489     ///
   490     /// \brief Calculates a minimum cut with \f$ source \f$ on the
   491     /// source-side (i.e. a set \f$ X\subsetneq V \f$ with \f$ source \in X \f$
   492     ///  and minimal out-degree).
   493     void calculateOut() {
   494       for (NodeIt it(*_graph); it != INVALID; ++it) {
   495         if (it != _source) {
   496           calculateOut(it);
   497           return;
   498         }
   499       }
   500     }
   501 
   502     /// \brief Calculates a minimum cut with \f$ source \f$ on the
   503     /// source-side.
   504     ///
   505     /// \brief Calculates a minimum cut with \f$ source \f$ on the
   506     /// source-side (i.e. a set \f$ X\subsetneq V \f$ with \f$ source \in X \f$
   507     ///  and minimal out-degree). The \c target is the initial target
   508     /// for the push-relabel algorithm.
   509     void calculateOut(const Node& target) {
   510       findMinCut(target, true, *_out_res_graph);
   511     }
   512 
   513     /// \brief Calculates a minimum cut with \f$ source \f$ on the
   514     /// sink-side.
   515     ///
   516     /// \brief Calculates a minimum cut with \f$ source \f$ on the
   517     /// sink-side (i.e. a set \f$ X\subsetneq V \f$ with 
   518     /// \f$ source \notin X \f$
   519     /// and minimal out-degree).
   520     void calculateIn() {
   521       for (NodeIt it(*_graph); it != INVALID; ++it) {
   522         if (it != _source) {
   523           calculateIn(it);
   524           return;
   525         }
   526       }
   527     }
   528 
   529     /// \brief Calculates a minimum cut with \f$ source \f$ on the
   530     /// sink-side.
   531     ///
   532     /// \brief Calculates a minimum cut with \f$ source \f$ on the
   533     /// sink-side (i.e. a set \f$ X\subsetneq V 
   534     /// \f$ with \f$ source \notin X \f$ and minimal out-degree).  
   535     /// The \c target is the initial
   536     /// target for the push-relabel algorithm.
   537     void calculateIn(const Node& target) {
   538       findMinCut(target, false, *_in_res_graph);
   539     }
   540 
   541     /// \brief Runs the algorithm.
   542     ///
   543     /// Runs the algorithm. It finds nodes \c source and \c target
   544     /// arbitrarily and then calls \ref init(), \ref calculateOut()
   545     /// and \ref calculateIn().
   546     void run() {
   547       init();
   548       for (NodeIt it(*_graph); it != INVALID; ++it) {
   549         if (it != _source) {
   550           calculateOut(it);
   551           calculateIn(it);
   552           return;
   553         }
   554       }
   555     }
   556 
   557     /// \brief Runs the algorithm.
   558     ///
   559     /// Runs the algorithm. It uses the given \c source node, finds a
   560     /// proper \c target and then calls the \ref init(), \ref
   561     /// calculateOut() and \ref calculateIn().
   562     void run(const Node& s) {
   563       init(s);
   564       for (NodeIt it(*_graph); it != INVALID; ++it) {
   565         if (it != _source) {
   566           calculateOut(it);
   567           calculateIn(it);
   568           return;
   569         }
   570       }
   571     }
   572 
   573     /// \brief Runs the algorithm.
   574     ///
   575     /// Runs the algorithm. It just calls the \ref init() and then
   576     /// \ref calculateOut() and \ref calculateIn().
   577     void run(const Node& s, const Node& t) {
   578       init(s); 
   579       calculateOut(t);
   580       calculateIn(t);
   581     }
   582 
   583     /// @}
   584     
   585     /// \name Query Functions 
   586     /// The result of the %HaoOrlin algorithm
   587     /// can be obtained using these functions.
   588     /// \n
   589     /// Before using these functions, either \ref run(), \ref
   590     /// calculateOut() or \ref calculateIn() must be called.
   591     
   592     /// @{
   593 
   594     /// \brief Returns the value of the minimum value cut.
   595     /// 
   596     /// Returns the value of the minimum value cut.
   597     Value minCut() const {
   598       return _min_cut;
   599     }
   600 
   601 
   602     /// \brief Returns a minimum cut.
   603     ///
   604     /// Sets \c nodeMap to the characteristic vector of a minimum
   605     /// value cut: it will give a nonempty set \f$ X\subsetneq V \f$
   606     /// with minimal out-degree (i.e. \c nodeMap will be true exactly
   607     /// for the nodes of \f$ X \f$).  \pre nodeMap should be a
   608     /// bool-valued node-map.
   609     template <typename NodeMap>
   610     Value minCut(NodeMap& nodeMap) const {
   611       for (NodeIt it(*_graph); it != INVALID; ++it) {
   612 	nodeMap.set(it, (*_min_cut_map)[it]);
   613       }
   614       return minCut();
   615     }
   616 
   617     /// @}
   618     
   619   }; //class HaoOrlin 
   620 
   621 
   622 } //namespace lemon
   623 
   624 #endif //LEMON_HAO_ORLIN_H