lemon/howard_mmc.h
author Peter Kovacs <kpeter@inf.elte.hu>
Mon, 16 Jul 2018 16:21:40 +0200
changeset 1183 cd72eae05bdf
parent 864 d3ea191c3412
child 1002 f63ba40a60f4
permissions -rw-r--r--
Change typenames to avoid Windows-specific compile issue (#612)
alpar@877
     1
/* -*- mode: C++; indent-tabs-mode: nil; -*-
kpeter@758
     2
 *
alpar@877
     3
 * This file is a part of LEMON, a generic C++ optimization library.
kpeter@758
     4
 *
alpar@877
     5
 * Copyright (C) 2003-2010
kpeter@758
     6
 * Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
kpeter@758
     7
 * (Egervary Research Group on Combinatorial Optimization, EGRES).
kpeter@758
     8
 *
kpeter@758
     9
 * Permission to use, modify and distribute this software is granted
kpeter@758
    10
 * provided that this copyright notice appears in all copies. For
kpeter@758
    11
 * precise terms see the accompanying LICENSE file.
kpeter@758
    12
 *
kpeter@758
    13
 * This software is provided "AS IS" with no warranty of any kind,
kpeter@758
    14
 * express or implied, and with no claim as to its suitability for any
kpeter@758
    15
 * purpose.
kpeter@758
    16
 *
kpeter@758
    17
 */
kpeter@758
    18
kpeter@864
    19
#ifndef LEMON_HOWARD_MMC_H
kpeter@864
    20
#define LEMON_HOWARD_MMC_H
kpeter@758
    21
kpeter@768
    22
/// \ingroup min_mean_cycle
kpeter@758
    23
///
kpeter@758
    24
/// \file
kpeter@758
    25
/// \brief Howard's algorithm for finding a minimum mean cycle.
kpeter@758
    26
kpeter@758
    27
#include <vector>
kpeter@763
    28
#include <limits>
kpeter@758
    29
#include <lemon/core.h>
kpeter@758
    30
#include <lemon/path.h>
kpeter@758
    31
#include <lemon/tolerance.h>
kpeter@758
    32
#include <lemon/connectivity.h>
kpeter@758
    33
kpeter@758
    34
namespace lemon {
kpeter@758
    35
kpeter@864
    36
  /// \brief Default traits class of HowardMmc class.
kpeter@761
    37
  ///
kpeter@864
    38
  /// Default traits class of HowardMmc class.
kpeter@761
    39
  /// \tparam GR The type of the digraph.
kpeter@864
    40
  /// \tparam CM The type of the cost map.
kpeter@761
    41
  /// It must conform to the \ref concepts::ReadMap "ReadMap" concept.
kpeter@761
    42
#ifdef DOXYGEN
kpeter@864
    43
  template <typename GR, typename CM>
kpeter@761
    44
#else
kpeter@864
    45
  template <typename GR, typename CM,
kpeter@864
    46
    bool integer = std::numeric_limits<typename CM::Value>::is_integer>
kpeter@761
    47
#endif
kpeter@864
    48
  struct HowardMmcDefaultTraits
kpeter@761
    49
  {
kpeter@761
    50
    /// The type of the digraph
kpeter@761
    51
    typedef GR Digraph;
kpeter@864
    52
    /// The type of the cost map
kpeter@864
    53
    typedef CM CostMap;
kpeter@864
    54
    /// The type of the arc costs
kpeter@864
    55
    typedef typename CostMap::Value Cost;
kpeter@761
    56
kpeter@864
    57
    /// \brief The large cost type used for internal computations
kpeter@761
    58
    ///
kpeter@864
    59
    /// The large cost type used for internal computations.
kpeter@864
    60
    /// It is \c long \c long if the \c Cost type is integer,
kpeter@761
    61
    /// otherwise it is \c double.
kpeter@864
    62
    /// \c Cost must be convertible to \c LargeCost.
kpeter@864
    63
    typedef double LargeCost;
kpeter@761
    64
kpeter@761
    65
    /// The tolerance type used for internal computations
kpeter@864
    66
    typedef lemon::Tolerance<LargeCost> Tolerance;
kpeter@761
    67
kpeter@761
    68
    /// \brief The path type of the found cycles
kpeter@761
    69
    ///
kpeter@761
    70
    /// The path type of the found cycles.
kpeter@761
    71
    /// It must conform to the \ref lemon::concepts::Path "Path" concept
kpeter@761
    72
    /// and it must have an \c addBack() function.
kpeter@761
    73
    typedef lemon::Path<Digraph> Path;
kpeter@761
    74
  };
kpeter@761
    75
kpeter@864
    76
  // Default traits class for integer cost types
kpeter@864
    77
  template <typename GR, typename CM>
kpeter@864
    78
  struct HowardMmcDefaultTraits<GR, CM, true>
kpeter@761
    79
  {
kpeter@761
    80
    typedef GR Digraph;
kpeter@864
    81
    typedef CM CostMap;
kpeter@864
    82
    typedef typename CostMap::Value Cost;
kpeter@761
    83
#ifdef LEMON_HAVE_LONG_LONG
kpeter@864
    84
    typedef long long LargeCost;
kpeter@761
    85
#else
kpeter@864
    86
    typedef long LargeCost;
kpeter@761
    87
#endif
kpeter@864
    88
    typedef lemon::Tolerance<LargeCost> Tolerance;
kpeter@761
    89
    typedef lemon::Path<Digraph> Path;
kpeter@761
    90
  };
kpeter@761
    91
kpeter@761
    92
kpeter@768
    93
  /// \addtogroup min_mean_cycle
kpeter@758
    94
  /// @{
kpeter@758
    95
kpeter@758
    96
  /// \brief Implementation of Howard's algorithm for finding a minimum
kpeter@758
    97
  /// mean cycle.
kpeter@758
    98
  ///
kpeter@764
    99
  /// This class implements Howard's policy iteration algorithm for finding
kpeter@864
   100
  /// a directed cycle of minimum mean cost in a digraph
kpeter@771
   101
  /// \ref amo93networkflows, \ref dasdan98minmeancycle.
kpeter@768
   102
  /// This class provides the most efficient algorithm for the
kpeter@768
   103
  /// minimum mean cycle problem, though the best known theoretical
kpeter@768
   104
  /// bound on its running time is exponential.
kpeter@758
   105
  ///
kpeter@758
   106
  /// \tparam GR The type of the digraph the algorithm runs on.
kpeter@864
   107
  /// \tparam CM The type of the cost map. The default
kpeter@758
   108
  /// map type is \ref concepts::Digraph::ArcMap "GR::ArcMap<int>".
kpeter@825
   109
  /// \tparam TR The traits class that defines various types used by the
kpeter@864
   110
  /// algorithm. By default, it is \ref HowardMmcDefaultTraits
kpeter@864
   111
  /// "HowardMmcDefaultTraits<GR, CM>".
kpeter@825
   112
  /// In most cases, this parameter should not be set directly,
kpeter@825
   113
  /// consider to use the named template parameters instead.
kpeter@758
   114
#ifdef DOXYGEN
kpeter@864
   115
  template <typename GR, typename CM, typename TR>
kpeter@758
   116
#else
kpeter@758
   117
  template < typename GR,
kpeter@864
   118
             typename CM = typename GR::template ArcMap<int>,
kpeter@864
   119
             typename TR = HowardMmcDefaultTraits<GR, CM> >
kpeter@758
   120
#endif
kpeter@864
   121
  class HowardMmc
kpeter@758
   122
  {
kpeter@758
   123
  public:
alpar@877
   124
kpeter@761
   125
    /// The type of the digraph
kpeter@761
   126
    typedef typename TR::Digraph Digraph;
kpeter@864
   127
    /// The type of the cost map
kpeter@864
   128
    typedef typename TR::CostMap CostMap;
kpeter@864
   129
    /// The type of the arc costs
kpeter@864
   130
    typedef typename TR::Cost Cost;
kpeter@761
   131
kpeter@864
   132
    /// \brief The large cost type
kpeter@761
   133
    ///
kpeter@864
   134
    /// The large cost type used for internal computations.
kpeter@864
   135
    /// By default, it is \c long \c long if the \c Cost type is integer,
kpeter@761
   136
    /// otherwise it is \c double.
kpeter@864
   137
    typedef typename TR::LargeCost LargeCost;
kpeter@761
   138
kpeter@761
   139
    /// The tolerance type
kpeter@761
   140
    typedef typename TR::Tolerance Tolerance;
kpeter@761
   141
kpeter@761
   142
    /// \brief The path type of the found cycles
kpeter@761
   143
    ///
kpeter@761
   144
    /// The path type of the found cycles.
kpeter@864
   145
    /// Using the \ref HowardMmcDefaultTraits "default traits class",
kpeter@761
   146
    /// it is \ref lemon::Path "Path<Digraph>".
kpeter@761
   147
    typedef typename TR::Path Path;
kpeter@761
   148
kpeter@864
   149
    /// The \ref HowardMmcDefaultTraits "traits class" of the algorithm
kpeter@761
   150
    typedef TR Traits;
kpeter@758
   151
kpeter@758
   152
  private:
kpeter@758
   153
kpeter@758
   154
    TEMPLATE_DIGRAPH_TYPEDEFS(Digraph);
alpar@877
   155
kpeter@758
   156
    // The digraph the algorithm runs on
kpeter@758
   157
    const Digraph &_gr;
kpeter@864
   158
    // The cost of the arcs
kpeter@864
   159
    const CostMap &_cost;
kpeter@758
   160
kpeter@760
   161
    // Data for the found cycles
kpeter@760
   162
    bool _curr_found, _best_found;
kpeter@864
   163
    LargeCost _curr_cost, _best_cost;
kpeter@760
   164
    int _curr_size, _best_size;
kpeter@760
   165
    Node _curr_node, _best_node;
kpeter@760
   166
kpeter@758
   167
    Path *_cycle_path;
kpeter@760
   168
    bool _local_path;
kpeter@758
   169
kpeter@760
   170
    // Internal data used by the algorithm
kpeter@760
   171
    typename Digraph::template NodeMap<Arc> _policy;
kpeter@760
   172
    typename Digraph::template NodeMap<bool> _reached;
kpeter@760
   173
    typename Digraph::template NodeMap<int> _level;
kpeter@864
   174
    typename Digraph::template NodeMap<LargeCost> _dist;
kpeter@758
   175
kpeter@760
   176
    // Data for storing the strongly connected components
kpeter@760
   177
    int _comp_num;
kpeter@758
   178
    typename Digraph::template NodeMap<int> _comp;
kpeter@760
   179
    std::vector<std::vector<Node> > _comp_nodes;
kpeter@760
   180
    std::vector<Node>* _nodes;
kpeter@760
   181
    typename Digraph::template NodeMap<std::vector<Arc> > _in_arcs;
alpar@877
   182
kpeter@760
   183
    // Queue used for BFS search
kpeter@760
   184
    std::vector<Node> _queue;
kpeter@760
   185
    int _qfront, _qback;
kpeter@761
   186
kpeter@761
   187
    Tolerance _tolerance;
alpar@877
   188
kpeter@767
   189
    // Infinite constant
kpeter@864
   190
    const LargeCost INF;
kpeter@767
   191
kpeter@761
   192
  public:
alpar@877
   193
kpeter@761
   194
    /// \name Named Template Parameters
kpeter@761
   195
    /// @{
kpeter@761
   196
kpeter@761
   197
    template <typename T>
kpeter@864
   198
    struct SetLargeCostTraits : public Traits {
kpeter@864
   199
      typedef T LargeCost;
kpeter@761
   200
      typedef lemon::Tolerance<T> Tolerance;
kpeter@761
   201
    };
kpeter@761
   202
kpeter@761
   203
    /// \brief \ref named-templ-param "Named parameter" for setting
kpeter@864
   204
    /// \c LargeCost type.
kpeter@761
   205
    ///
kpeter@864
   206
    /// \ref named-templ-param "Named parameter" for setting \c LargeCost
kpeter@761
   207
    /// type. It is used for internal computations in the algorithm.
kpeter@761
   208
    template <typename T>
kpeter@864
   209
    struct SetLargeCost
kpeter@864
   210
      : public HowardMmc<GR, CM, SetLargeCostTraits<T> > {
kpeter@864
   211
      typedef HowardMmc<GR, CM, SetLargeCostTraits<T> > Create;
kpeter@761
   212
    };
kpeter@761
   213
kpeter@761
   214
    template <typename T>
kpeter@761
   215
    struct SetPathTraits : public Traits {
kpeter@761
   216
      typedef T Path;
kpeter@761
   217
    };
kpeter@761
   218
kpeter@761
   219
    /// \brief \ref named-templ-param "Named parameter" for setting
kpeter@761
   220
    /// \c %Path type.
kpeter@761
   221
    ///
kpeter@761
   222
    /// \ref named-templ-param "Named parameter" for setting the \c %Path
kpeter@761
   223
    /// type of the found cycles.
kpeter@761
   224
    /// It must conform to the \ref lemon::concepts::Path "Path" concept
kpeter@761
   225
    /// and it must have an \c addBack() function.
kpeter@761
   226
    template <typename T>
kpeter@761
   227
    struct SetPath
kpeter@864
   228
      : public HowardMmc<GR, CM, SetPathTraits<T> > {
kpeter@864
   229
      typedef HowardMmc<GR, CM, SetPathTraits<T> > Create;
kpeter@761
   230
    };
alpar@877
   231
kpeter@761
   232
    /// @}
kpeter@758
   233
kpeter@863
   234
  protected:
kpeter@863
   235
kpeter@864
   236
    HowardMmc() {}
kpeter@863
   237
kpeter@758
   238
  public:
kpeter@758
   239
kpeter@758
   240
    /// \brief Constructor.
kpeter@758
   241
    ///
kpeter@758
   242
    /// The constructor of the class.
kpeter@758
   243
    ///
kpeter@758
   244
    /// \param digraph The digraph the algorithm runs on.
kpeter@864
   245
    /// \param cost The costs of the arcs.
kpeter@864
   246
    HowardMmc( const Digraph &digraph,
kpeter@864
   247
               const CostMap &cost ) :
kpeter@864
   248
      _gr(digraph), _cost(cost), _best_found(false),
kpeter@864
   249
      _best_cost(0), _best_size(1), _cycle_path(NULL), _local_path(false),
kpeter@760
   250
      _policy(digraph), _reached(digraph), _level(digraph), _dist(digraph),
kpeter@767
   251
      _comp(digraph), _in_arcs(digraph),
kpeter@864
   252
      INF(std::numeric_limits<LargeCost>::has_infinity ?
kpeter@864
   253
          std::numeric_limits<LargeCost>::infinity() :
kpeter@864
   254
          std::numeric_limits<LargeCost>::max())
kpeter@758
   255
    {}
kpeter@758
   256
kpeter@758
   257
    /// Destructor.
kpeter@864
   258
    ~HowardMmc() {
kpeter@758
   259
      if (_local_path) delete _cycle_path;
kpeter@758
   260
    }
kpeter@758
   261
kpeter@758
   262
    /// \brief Set the path structure for storing the found cycle.
kpeter@758
   263
    ///
kpeter@758
   264
    /// This function sets an external path structure for storing the
kpeter@758
   265
    /// found cycle.
kpeter@758
   266
    ///
kpeter@758
   267
    /// If you don't call this function before calling \ref run() or
kpeter@864
   268
    /// \ref findCycleMean(), it will allocate a local \ref Path "path"
kpeter@758
   269
    /// structure. The destuctor deallocates this automatically
kpeter@758
   270
    /// allocated object, of course.
kpeter@758
   271
    ///
kpeter@758
   272
    /// \note The algorithm calls only the \ref lemon::Path::addBack()
kpeter@758
   273
    /// "addBack()" function of the given path structure.
kpeter@758
   274
    ///
kpeter@758
   275
    /// \return <tt>(*this)</tt>
kpeter@864
   276
    HowardMmc& cycle(Path &path) {
kpeter@758
   277
      if (_local_path) {
kpeter@758
   278
        delete _cycle_path;
kpeter@758
   279
        _local_path = false;
kpeter@758
   280
      }
kpeter@758
   281
      _cycle_path = &path;
kpeter@758
   282
      return *this;
kpeter@758
   283
    }
kpeter@758
   284
kpeter@769
   285
    /// \brief Set the tolerance used by the algorithm.
kpeter@769
   286
    ///
kpeter@769
   287
    /// This function sets the tolerance object used by the algorithm.
kpeter@769
   288
    ///
kpeter@769
   289
    /// \return <tt>(*this)</tt>
kpeter@864
   290
    HowardMmc& tolerance(const Tolerance& tolerance) {
kpeter@769
   291
      _tolerance = tolerance;
kpeter@769
   292
      return *this;
kpeter@769
   293
    }
kpeter@769
   294
kpeter@769
   295
    /// \brief Return a const reference to the tolerance.
kpeter@769
   296
    ///
kpeter@769
   297
    /// This function returns a const reference to the tolerance object
kpeter@769
   298
    /// used by the algorithm.
kpeter@769
   299
    const Tolerance& tolerance() const {
kpeter@769
   300
      return _tolerance;
kpeter@769
   301
    }
kpeter@769
   302
kpeter@758
   303
    /// \name Execution control
kpeter@758
   304
    /// The simplest way to execute the algorithm is to call the \ref run()
kpeter@758
   305
    /// function.\n
kpeter@864
   306
    /// If you only need the minimum mean cost, you may call
kpeter@864
   307
    /// \ref findCycleMean().
kpeter@758
   308
kpeter@758
   309
    /// @{
kpeter@758
   310
kpeter@758
   311
    /// \brief Run the algorithm.
kpeter@758
   312
    ///
kpeter@758
   313
    /// This function runs the algorithm.
kpeter@759
   314
    /// It can be called more than once (e.g. if the underlying digraph
kpeter@864
   315
    /// and/or the arc costs have been modified).
kpeter@758
   316
    ///
kpeter@758
   317
    /// \return \c true if a directed cycle exists in the digraph.
kpeter@758
   318
    ///
kpeter@759
   319
    /// \note <tt>mmc.run()</tt> is just a shortcut of the following code.
kpeter@758
   320
    /// \code
kpeter@864
   321
    ///   return mmc.findCycleMean() && mmc.findCycle();
kpeter@758
   322
    /// \endcode
kpeter@758
   323
    bool run() {
kpeter@864
   324
      return findCycleMean() && findCycle();
kpeter@758
   325
    }
kpeter@758
   326
kpeter@759
   327
    /// \brief Find the minimum cycle mean.
kpeter@758
   328
    ///
kpeter@864
   329
    /// This function finds the minimum mean cost of the directed
kpeter@759
   330
    /// cycles in the digraph.
kpeter@758
   331
    ///
kpeter@759
   332
    /// \return \c true if a directed cycle exists in the digraph.
kpeter@864
   333
    bool findCycleMean() {
kpeter@760
   334
      // Initialize and find strongly connected components
kpeter@760
   335
      init();
kpeter@760
   336
      findComponents();
alpar@877
   337
kpeter@759
   338
      // Find the minimum cycle mean in the components
kpeter@758
   339
      for (int comp = 0; comp < _comp_num; ++comp) {
kpeter@760
   340
        // Find the minimum mean cycle in the current component
kpeter@760
   341
        if (!buildPolicyGraph(comp)) continue;
kpeter@758
   342
        while (true) {
kpeter@760
   343
          findPolicyCycle();
kpeter@758
   344
          if (!computeNodeDistances()) break;
kpeter@758
   345
        }
kpeter@760
   346
        // Update the best cycle (global minimum mean cycle)
kpeter@767
   347
        if ( _curr_found && (!_best_found ||
kpeter@864
   348
             _curr_cost * _best_size < _best_cost * _curr_size) ) {
kpeter@760
   349
          _best_found = true;
kpeter@864
   350
          _best_cost = _curr_cost;
kpeter@760
   351
          _best_size = _curr_size;
kpeter@760
   352
          _best_node = _curr_node;
kpeter@760
   353
        }
kpeter@758
   354
      }
kpeter@760
   355
      return _best_found;
kpeter@758
   356
    }
kpeter@758
   357
kpeter@758
   358
    /// \brief Find a minimum mean directed cycle.
kpeter@758
   359
    ///
kpeter@864
   360
    /// This function finds a directed cycle of minimum mean cost
kpeter@864
   361
    /// in the digraph using the data computed by findCycleMean().
kpeter@758
   362
    ///
kpeter@758
   363
    /// \return \c true if a directed cycle exists in the digraph.
kpeter@758
   364
    ///
kpeter@864
   365
    /// \pre \ref findCycleMean() must be called before using this function.
kpeter@758
   366
    bool findCycle() {
kpeter@760
   367
      if (!_best_found) return false;
kpeter@760
   368
      _cycle_path->addBack(_policy[_best_node]);
kpeter@760
   369
      for ( Node v = _best_node;
kpeter@760
   370
            (v = _gr.target(_policy[v])) != _best_node; ) {
kpeter@758
   371
        _cycle_path->addBack(_policy[v]);
kpeter@758
   372
      }
kpeter@758
   373
      return true;
kpeter@758
   374
    }
kpeter@758
   375
kpeter@758
   376
    /// @}
kpeter@758
   377
kpeter@758
   378
    /// \name Query Functions
kpeter@759
   379
    /// The results of the algorithm can be obtained using these
kpeter@758
   380
    /// functions.\n
kpeter@758
   381
    /// The algorithm should be executed before using them.
kpeter@758
   382
kpeter@758
   383
    /// @{
kpeter@758
   384
kpeter@864
   385
    /// \brief Return the total cost of the found cycle.
kpeter@758
   386
    ///
kpeter@864
   387
    /// This function returns the total cost of the found cycle.
kpeter@758
   388
    ///
kpeter@864
   389
    /// \pre \ref run() or \ref findCycleMean() must be called before
kpeter@758
   390
    /// using this function.
kpeter@864
   391
    Cost cycleCost() const {
kpeter@864
   392
      return static_cast<Cost>(_best_cost);
kpeter@758
   393
    }
kpeter@758
   394
kpeter@758
   395
    /// \brief Return the number of arcs on the found cycle.
kpeter@758
   396
    ///
kpeter@758
   397
    /// This function returns the number of arcs on the found cycle.
kpeter@758
   398
    ///
kpeter@864
   399
    /// \pre \ref run() or \ref findCycleMean() must be called before
kpeter@758
   400
    /// using this function.
kpeter@864
   401
    int cycleSize() const {
kpeter@760
   402
      return _best_size;
kpeter@758
   403
    }
kpeter@758
   404
kpeter@864
   405
    /// \brief Return the mean cost of the found cycle.
kpeter@758
   406
    ///
kpeter@864
   407
    /// This function returns the mean cost of the found cycle.
kpeter@758
   408
    ///
kpeter@760
   409
    /// \note <tt>alg.cycleMean()</tt> is just a shortcut of the
kpeter@758
   410
    /// following code.
kpeter@758
   411
    /// \code
kpeter@864
   412
    ///   return static_cast<double>(alg.cycleCost()) / alg.cycleSize();
kpeter@758
   413
    /// \endcode
kpeter@758
   414
    ///
kpeter@864
   415
    /// \pre \ref run() or \ref findCycleMean() must be called before
kpeter@758
   416
    /// using this function.
kpeter@758
   417
    double cycleMean() const {
kpeter@864
   418
      return static_cast<double>(_best_cost) / _best_size;
kpeter@758
   419
    }
kpeter@758
   420
kpeter@758
   421
    /// \brief Return the found cycle.
kpeter@758
   422
    ///
kpeter@758
   423
    /// This function returns a const reference to the path structure
kpeter@758
   424
    /// storing the found cycle.
kpeter@758
   425
    ///
kpeter@758
   426
    /// \pre \ref run() or \ref findCycle() must be called before using
kpeter@758
   427
    /// this function.
kpeter@758
   428
    const Path& cycle() const {
kpeter@758
   429
      return *_cycle_path;
kpeter@758
   430
    }
kpeter@758
   431
kpeter@758
   432
    ///@}
kpeter@758
   433
kpeter@758
   434
  private:
kpeter@758
   435
kpeter@760
   436
    // Initialize
kpeter@760
   437
    void init() {
kpeter@760
   438
      if (!_cycle_path) {
kpeter@760
   439
        _local_path = true;
kpeter@760
   440
        _cycle_path = new Path;
kpeter@758
   441
      }
kpeter@760
   442
      _queue.resize(countNodes(_gr));
kpeter@760
   443
      _best_found = false;
kpeter@864
   444
      _best_cost = 0;
kpeter@760
   445
      _best_size = 1;
kpeter@760
   446
      _cycle_path->clear();
kpeter@760
   447
    }
alpar@877
   448
kpeter@760
   449
    // Find strongly connected components and initialize _comp_nodes
kpeter@760
   450
    // and _in_arcs
kpeter@760
   451
    void findComponents() {
kpeter@760
   452
      _comp_num = stronglyConnectedComponents(_gr, _comp);
kpeter@760
   453
      _comp_nodes.resize(_comp_num);
kpeter@760
   454
      if (_comp_num == 1) {
kpeter@760
   455
        _comp_nodes[0].clear();
kpeter@760
   456
        for (NodeIt n(_gr); n != INVALID; ++n) {
kpeter@760
   457
          _comp_nodes[0].push_back(n);
kpeter@760
   458
          _in_arcs[n].clear();
kpeter@760
   459
          for (InArcIt a(_gr, n); a != INVALID; ++a) {
kpeter@760
   460
            _in_arcs[n].push_back(a);
kpeter@760
   461
          }
kpeter@760
   462
        }
kpeter@760
   463
      } else {
kpeter@760
   464
        for (int i = 0; i < _comp_num; ++i)
kpeter@760
   465
          _comp_nodes[i].clear();
kpeter@760
   466
        for (NodeIt n(_gr); n != INVALID; ++n) {
kpeter@760
   467
          int k = _comp[n];
kpeter@760
   468
          _comp_nodes[k].push_back(n);
kpeter@760
   469
          _in_arcs[n].clear();
kpeter@760
   470
          for (InArcIt a(_gr, n); a != INVALID; ++a) {
kpeter@760
   471
            if (_comp[_gr.source(a)] == k) _in_arcs[n].push_back(a);
kpeter@760
   472
          }
kpeter@760
   473
        }
kpeter@758
   474
      }
kpeter@760
   475
    }
kpeter@760
   476
kpeter@760
   477
    // Build the policy graph in the given strongly connected component
kpeter@760
   478
    // (the out-degree of every node is 1)
kpeter@760
   479
    bool buildPolicyGraph(int comp) {
kpeter@760
   480
      _nodes = &(_comp_nodes[comp]);
kpeter@760
   481
      if (_nodes->size() < 1 ||
kpeter@760
   482
          (_nodes->size() == 1 && _in_arcs[(*_nodes)[0]].size() == 0)) {
kpeter@760
   483
        return false;
kpeter@758
   484
      }
kpeter@760
   485
      for (int i = 0; i < int(_nodes->size()); ++i) {
kpeter@767
   486
        _dist[(*_nodes)[i]] = INF;
kpeter@760
   487
      }
kpeter@760
   488
      Node u, v;
kpeter@760
   489
      Arc e;
kpeter@760
   490
      for (int i = 0; i < int(_nodes->size()); ++i) {
kpeter@760
   491
        v = (*_nodes)[i];
kpeter@760
   492
        for (int j = 0; j < int(_in_arcs[v].size()); ++j) {
kpeter@760
   493
          e = _in_arcs[v][j];
kpeter@760
   494
          u = _gr.source(e);
kpeter@864
   495
          if (_cost[e] < _dist[u]) {
kpeter@864
   496
            _dist[u] = _cost[e];
kpeter@760
   497
            _policy[u] = e;
kpeter@760
   498
          }
kpeter@758
   499
        }
kpeter@758
   500
      }
kpeter@758
   501
      return true;
kpeter@758
   502
    }
kpeter@758
   503
kpeter@760
   504
    // Find the minimum mean cycle in the policy graph
kpeter@760
   505
    void findPolicyCycle() {
kpeter@760
   506
      for (int i = 0; i < int(_nodes->size()); ++i) {
kpeter@760
   507
        _level[(*_nodes)[i]] = -1;
kpeter@760
   508
      }
kpeter@864
   509
      LargeCost ccost;
kpeter@758
   510
      int csize;
kpeter@758
   511
      Node u, v;
kpeter@760
   512
      _curr_found = false;
kpeter@760
   513
      for (int i = 0; i < int(_nodes->size()); ++i) {
kpeter@760
   514
        u = (*_nodes)[i];
kpeter@760
   515
        if (_level[u] >= 0) continue;
kpeter@760
   516
        for (; _level[u] < 0; u = _gr.target(_policy[u])) {
kpeter@760
   517
          _level[u] = i;
kpeter@760
   518
        }
kpeter@760
   519
        if (_level[u] == i) {
kpeter@760
   520
          // A cycle is found
kpeter@864
   521
          ccost = _cost[_policy[u]];
kpeter@760
   522
          csize = 1;
kpeter@760
   523
          for (v = u; (v = _gr.target(_policy[v])) != u; ) {
kpeter@864
   524
            ccost += _cost[_policy[v]];
kpeter@760
   525
            ++csize;
kpeter@758
   526
          }
kpeter@760
   527
          if ( !_curr_found ||
kpeter@864
   528
               (ccost * _curr_size < _curr_cost * csize) ) {
kpeter@760
   529
            _curr_found = true;
kpeter@864
   530
            _curr_cost = ccost;
kpeter@760
   531
            _curr_size = csize;
kpeter@760
   532
            _curr_node = u;
kpeter@758
   533
          }
kpeter@758
   534
        }
kpeter@758
   535
      }
kpeter@758
   536
    }
kpeter@758
   537
kpeter@760
   538
    // Contract the policy graph and compute node distances
kpeter@758
   539
    bool computeNodeDistances() {
kpeter@760
   540
      // Find the component of the main cycle and compute node distances
kpeter@760
   541
      // using reverse BFS
kpeter@760
   542
      for (int i = 0; i < int(_nodes->size()); ++i) {
kpeter@760
   543
        _reached[(*_nodes)[i]] = false;
kpeter@760
   544
      }
kpeter@760
   545
      _qfront = _qback = 0;
kpeter@760
   546
      _queue[0] = _curr_node;
kpeter@760
   547
      _reached[_curr_node] = true;
kpeter@760
   548
      _dist[_curr_node] = 0;
kpeter@758
   549
      Node u, v;
kpeter@760
   550
      Arc e;
kpeter@760
   551
      while (_qfront <= _qback) {
kpeter@760
   552
        v = _queue[_qfront++];
kpeter@760
   553
        for (int j = 0; j < int(_in_arcs[v].size()); ++j) {
kpeter@760
   554
          e = _in_arcs[v][j];
kpeter@758
   555
          u = _gr.source(e);
kpeter@760
   556
          if (_policy[u] == e && !_reached[u]) {
kpeter@760
   557
            _reached[u] = true;
kpeter@864
   558
            _dist[u] = _dist[v] + _cost[e] * _curr_size - _curr_cost;
kpeter@760
   559
            _queue[++_qback] = u;
kpeter@758
   560
          }
kpeter@758
   561
        }
kpeter@758
   562
      }
kpeter@760
   563
kpeter@760
   564
      // Connect all other nodes to this component and compute node
kpeter@760
   565
      // distances using reverse BFS
kpeter@760
   566
      _qfront = 0;
kpeter@760
   567
      while (_qback < int(_nodes->size())-1) {
kpeter@760
   568
        v = _queue[_qfront++];
kpeter@760
   569
        for (int j = 0; j < int(_in_arcs[v].size()); ++j) {
kpeter@760
   570
          e = _in_arcs[v][j];
kpeter@760
   571
          u = _gr.source(e);
kpeter@760
   572
          if (!_reached[u]) {
kpeter@760
   573
            _reached[u] = true;
kpeter@760
   574
            _policy[u] = e;
kpeter@864
   575
            _dist[u] = _dist[v] + _cost[e] * _curr_size - _curr_cost;
kpeter@760
   576
            _queue[++_qback] = u;
kpeter@760
   577
          }
kpeter@760
   578
        }
kpeter@760
   579
      }
kpeter@760
   580
kpeter@760
   581
      // Improve node distances
kpeter@758
   582
      bool improved = false;
kpeter@760
   583
      for (int i = 0; i < int(_nodes->size()); ++i) {
kpeter@760
   584
        v = (*_nodes)[i];
kpeter@760
   585
        for (int j = 0; j < int(_in_arcs[v].size()); ++j) {
kpeter@760
   586
          e = _in_arcs[v][j];
kpeter@760
   587
          u = _gr.source(e);
kpeter@864
   588
          LargeCost delta = _dist[v] + _cost[e] * _curr_size - _curr_cost;
kpeter@761
   589
          if (_tolerance.less(delta, _dist[u])) {
kpeter@760
   590
            _dist[u] = delta;
kpeter@760
   591
            _policy[u] = e;
kpeter@760
   592
            improved = true;
kpeter@760
   593
          }
kpeter@758
   594
        }
kpeter@758
   595
      }
kpeter@758
   596
      return improved;
kpeter@758
   597
    }
kpeter@758
   598
kpeter@864
   599
  }; //class HowardMmc
kpeter@758
   600
kpeter@758
   601
  ///@}
kpeter@758
   602
kpeter@758
   603
} //namespace lemon
kpeter@758
   604
kpeter@864
   605
#endif //LEMON_HOWARD_MMC_H