lemon/tabu_search.h
author deba
Sun, 30 Dec 2007 18:23:32 +0000
changeset 2550 f26368148b9c
parent 2370 ed6539025f27
child 2553 bfced05fa852
permissions -rw-r--r--
Changing degree of tournament tree
Bug fix in union find
Small efficiency improvment in bipartite matchings
     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_TABU_SEARCH_H
    20 #define LEMON_TABU_SEARCH_H
    21 
    22 /// \ingroup metah
    23 /// \file
    24 /// \brief TabuSearch algorithm.
    25 ///
    26 /// \author Szabadkai Mark
    27 
    28 #include <lemon/bits/utility.h>
    29 #include <lemon/error.h>
    30 #include <lemon/time_measure.h>
    31 #include <functional>
    32 #include <deque>
    33 
    34 
    35 namespace lemon {
    36 
    37   /// \brief Default Traits for TabuSearch class.
    38   /// 
    39   /// This template defines the needed types for the \ref TabuSearch class.
    40   /// Is main purpos is to simplify the main class's template interface,
    41   /// but it provides the EdgeIt type, passing to the concrete graph wheter
    42   /// it is directed or undirected.
    43 #ifdef DOXYGEN
    44   template< typename GRAPH, typename VALUE, 
    45             typename HEIGHTMAP, typename BETTER, bool UNDIR >
    46 #else
    47   template< typename GRAPH, typename VALUE,
    48             typename HEIGHTMAP = typename GRAPH::template NodeMap<VALUE>,
    49             typename BETTER = std::less<VALUE>,
    50             bool UNDIR = UndirectedTagIndicator<GRAPH>::value >
    51 #endif
    52   struct TabuSearchDefaultTraits {
    53     typedef  VALUE  Value; 
    54     typedef  BETTER  Better;
    55 
    56     typedef  GRAPH  Graph;
    57     typedef  typename GRAPH::Node  Node;
    58     typedef  HEIGHTMAP  HeightMap;
    59 
    60     typedef  typename GRAPH::IncEdgeIt  EdgeIt;
    61   };
    62 
    63   template< typename GRAPH, typename VALUE, 
    64             typename HEIGHTMAP, typename BETTER >
    65   struct TabuSearchDefaultTraits< GRAPH, VALUE, HEIGHTMAP, BETTER, false > {
    66     typedef  VALUE  Value;
    67     typedef  BETTER  Better;
    68 
    69     typedef  GRAPH  Graph;
    70     typedef  typename GRAPH::Node  Node;
    71     typedef  HEIGHTMAP  HeightMap;
    72 
    73     typedef  typename GRAPH::OutEdgeIt  EdgeIt;
    74   };
    75 
    76 
    77 
    78   /// \brief Policy hierarchy to controll the search algorithm.
    79   ///
    80   /// The fallowing template hierarchy offers a clean interface to define own
    81   /// policies, and combine existing ones.
    82   template< typename TS >
    83   struct TabuSearchPolicyConcept {
    84     void  target( TS *ts ) {}
    85 
    86     void  reset()  {}
    87     bool  onStep() { return false; }
    88     bool  onStick() { return false; }
    89     bool  onImprove( const typename TS::Value &best ) { return false; }
    90   };
    91 
    92   template< typename TS >
    93   struct YesPolicy {
    94     void  target( TS *ts ) {}
    95 
    96     void  reset()  {}
    97     bool  onStep() { return true; }
    98     bool  onStick() { return true; }
    99     bool  onImprove( const typename TS::Value &best ) { return true; }
   100   };
   101 
   102   template< typename TS >
   103   struct NoPolicy : public TabuSearchPolicyConcept<TS> {};
   104 
   105   /// \brief Some basic methode, how tow Policies can be combined
   106   struct PolicyAndCombination {
   107     static bool  evaluate( const bool r1, const bool r2 ) {
   108       return r1 && r2;
   109     }
   110   };
   111 
   112   struct PolicyOrCombination {
   113     static bool  evaluate( const bool r1, const bool r2 ) {
   114       return r1 || r2;
   115     }
   116   };
   117 
   118   /// \brief CombinePolicies
   119   ///
   120   /// It combines tow policies using the given combination methode (mainly
   121   /// some of the basic logical methodes) to create a new one.
   122 #ifdef DOXYGEN
   123   template< template<typename> class CP1, template<typename> class CP2, 
   124             typename COMBINATION >
   125 #else
   126   template< template<typename> class CP1, template<typename> class CP2,
   127             typename COMBINATION = PolicyAndCombination >
   128 #endif
   129   struct CombinePolicies {
   130     template< typename TS >
   131     struct Policy {
   132       typedef CP1<TS>  Policy1;
   133       typedef CP2<TS>  Policy2;
   134       
   135       Policy1  policy1;
   136       Policy2  policy2;
   137 
   138       inline Policy() : policy1(), policy2() {}
   139       inline Policy( const Policy1 &cp1, const Policy2 &cp2 ) 
   140         : policy1(cp1), policy2(cp2) {}
   141 
   142       void  target( TS *ts ) {
   143         policy1.target(ts), policy2.target(ts);
   144       };
   145 
   146       void  reset() {
   147         policy1.reset(), policy2.reset();
   148       }
   149 
   150       bool  onStep() {
   151         return cmb.evaluate( policy1.onStep(), policy2.onStep() );
   152       }
   153 
   154       bool  onStick() {
   155         return cmb.evaluate( policy1.onStick(), policy2.onStick() );
   156       }
   157 
   158       bool  onImprove( const typename TS::Value &best ) {
   159         return cmb.evaluate( policy1.onImprove(best), 
   160                              policy2.onImprove(best) );
   161       }
   162 
   163     private:
   164       COMBINATION cmb;
   165     };
   166   };
   167 
   168 
   169   /// \brief IterationPolicy limits the number of iterations and the
   170   /// number of iterations without improvement
   171   template< typename TS >
   172   struct IterationPolicy {
   173     IterationPolicy() : _it_lim(100000), _noimpr_it_lim(5000) {}
   174     IterationPolicy( const long int itl, const long int noimpritl )
   175       : _it_lim(itl), _noimpr_it_lim(noimpritl)
   176     {}
   177 
   178     void  target( TS *ts ) {}
   179 
   180     void  reset() {
   181       _it = _noimpr_it = 0;
   182     }
   183 
   184     bool  onStep() {
   185       ++_it; ++_noimpr_it;
   186       return (_it <= _it_lim) && (_noimpr_it <= _noimpr_it_lim);
   187     }
   188 		
   189     bool  onStick() {
   190       return false;
   191     }
   192 
   193     bool  onImprove( const typename TS::Value &best ) {
   194       _noimpr_it = 0;
   195       return true;
   196     }
   197 
   198     long int  iterationLimit() const {
   199       return _it_lim;
   200     }
   201 
   202     void  iterationLimit( const long int itl ) {
   203       _it_lim = itl;
   204     }
   205 
   206     long int  noImprovementIterationLimit() const {
   207       return _noimpr_it_lim;
   208     }
   209 
   210     void  noImprovementIterationLimit( const long int noimpritl ) {
   211       _noimpr_it_lim = noimpritl;
   212     }
   213 
   214   private:
   215     long int  _it_lim, _noimpr_it_lim;
   216     long int  _it, _noimpr_it;
   217   };
   218 
   219   /// \brief HeightPolicy stops the search when a given height is reached or
   220   /// exceeds
   221   template< typename TS >
   222   struct HeightPolicy {
   223     typedef typename TS::Value  Value;
   224 
   225     HeightPolicy() : _height_lim(), _found(false) {}
   226     HeightPolicy( const Value &hl ) : _height_lim(hl), _found(false) {}
   227 
   228     void  target( TS *ts ) {}
   229 
   230     void  reset() {
   231       _found = false;
   232     }
   233 
   234     bool  onStep() {
   235       return !_found;
   236     }
   237 
   238     bool  onStick() {
   239       return false;
   240     }
   241 
   242     bool  onImprove( const Value &best ) {
   243       typename TS::Better  better;
   244       _found = better(best, _height_lim) || (best == _height_lim);
   245       return !_found;
   246     }
   247 
   248     Value  heightLimi() const {
   249       return _height_lim;
   250     }
   251 
   252     void  heightLimi( const Value &hl ) {
   253       _height_lim = hl;
   254     }
   255 
   256   private:
   257     Value  _height_lim;
   258     bool  _found;
   259   };
   260 
   261   /// \brief TimePolicy limits the time for searching.
   262   template< typename TS >
   263   struct TimePolicy {
   264     TimePolicy() : _time_lim(60.0), _timeisup(false) {}
   265     TimePolicy( const double tl ) : _time_lim(tl), _timeisup(false) {}
   266 
   267     void  target( TS *ts ) {}
   268 
   269     void  reset() {
   270       _timeisup = false;
   271       _t.reset();
   272     }
   273 
   274     bool  onStep() {
   275       update();
   276       return !_timeisup;
   277     }
   278 
   279     bool  onStick() {
   280       return false;
   281     }
   282 
   283     bool  onImprove( const typename TS::Value &best ) {
   284       update();
   285       return !_timeisup;
   286     }
   287 
   288     double timeLimit() const {
   289       return _time_lim;
   290     }
   291 
   292     void  setTimeLimit( const double tl ) {
   293       _time_lim = tl;
   294       update();
   295     }
   296 
   297   private:
   298     lemon::Timer  _t;
   299     double  _time_lim;
   300     bool  _timeisup;
   301 
   302     inline void  update() {
   303       _timeisup = _t.realTime() > _time_lim;
   304     }
   305   };
   306 
   307 
   308 
   309   /// \ingroup metah
   310   ///
   311   /// \brief TabuSearch main class
   312   ///
   313   /// This class offers the implementation of tabu-search algorithm. The
   314   /// tabu-serach is a local-search. It starts from a specified point of the
   315   /// problem's graph representation, and in every step it goes to the localy
   316   /// best next Node except those in tabu set. The maximum size of this tabu
   317   /// set defines how many Node will be remembered. The best Node ever found
   318   /// will also stored, so we wont lose it, even is the search continues.
   319   /// The class can be used on any kind of Graph and with any kind of Value
   320   /// with a total-settlement on it.
   321   ///
   322   /// \param _Graph The graph type the algorithm runs on.
   323   /// \param _Value The values' type associated to the nodes.
   324   /// \param _Policy Controlls the search. Determinates when to stop, or how
   325   /// manage stuck search. Default value is \ref IterationPolicy .
   326   /// \param _Traits Collection of needed types. Default value is
   327   /// \ref TabuSearchDefaultTraits .
   328   ///
   329   /// \author Szabadkai Mark
   330 #ifdef DOXYGEN
   331   template< typename GRAPH, typename VALUE, template<typename> class POLICY, typename TRAITS >
   332 #else
   333   template< typename GRAPH, typename VALUE,
   334             template<typename> class POLICY = IterationPolicy,
   335             typename TRAITS = TabuSearchDefaultTraits<GRAPH, VALUE> >
   336 #endif
   337   class TabuSearch
   338   {
   339   public:
   340 
   341     /// \brief Thrown by setting the size of the tabu-set and the given size
   342     /// is less than 2.
   343     class BadParameterError : public lemon::LogicError {
   344     public:
   345       virtual const char* what() const throw() {
   346         return "lemon::TabuSearch::BadParameterError";
   347       }
   348     };
   349 
   350     ///Public types
   351     typedef  TabuSearch<GRAPH,VALUE,POLICY,TRAITS>  SelfType;
   352 
   353     typedef  typename TRAITS::Graph  Graph;
   354     typedef  typename TRAITS::Node  Node;
   355     typedef  typename TRAITS::Value  Value;
   356     typedef  typename TRAITS::HeightMap  HeightMap;
   357     typedef  typename TRAITS::Better  Better;
   358     typedef  typename std::deque< Node >::const_iterator  TabuIterator;
   359 
   360     typedef  POLICY<SelfType>  Policy;
   361 
   362   protected:
   363     typedef  typename TRAITS::EdgeIt  EdgeIt;
   364 
   365     const Graph  &gr;
   366     const HeightMap  &height;
   367     /// The tabu set. Teh current node is the first
   368     std::deque< Node >  tabu;
   369     /// Maximal tabu size
   370     unsigned int  mts;
   371     /// The best Node found
   372     Node  b;
   373 
   374     Better  better;
   375     Policy  pol;
   376 
   377   public:
   378     /// \brief Constructor
   379     ///
   380     /// \param graph the graph the algorithm will run on.
   381     /// \param hm the height map used by the algorithm.
   382     /// \param tabusz the maximal size of the tabu set. Default value is 3
   383     /// \param p the Policy controlling the search.
   384     TabuSearch( const Graph &graph, const HeightMap &hm, 
   385                 const int tabusz = 3, Policy p = Policy() )
   386       : gr(graph), height(hm), mts(tabusz), pol(p)
   387     {
   388       pol.target(this);
   389     }
   390 
   391     /// \brief Destructor
   392     ~TabuSearch() {
   393       pol.target(NULL);
   394     }
   395 
   396     /// Set/Get the size of the tabu set
   397     void  tabuSize( const unsigned int size )
   398     {
   399       if( size < 2 )
   400       throw BadParameterError( "Tabu size must be at least 2!" );
   401       mts = size;
   402       while( mts < tabu.size() )
   403       tabu.pop_back();
   404     }
   405 
   406     unsigned int  tabuSize() const {
   407       return mts;
   408     }
   409 
   410     /// Set/Get Policy
   411     void  policy( Policy p ) {
   412       pol.target(NULL);
   413       pol = p;
   414       pol.target(this);
   415     }
   416 		
   417     Policy& policy()  {
   418       return pol;
   419     }
   420 
   421     /// \name Execution control
   422     /// The simplest way to execute the algorithm is to use the member
   423     /// functions called \c run( 'startnode' ).
   424     ///@{
   425 
   426     /// \brief Initializes the internal data.
   427     ///
   428     /// \param startn The start node where the search begins.
   429     void  init( const Node startn ) {
   430       tabu.clear();
   431       tabu.push_front( startn );
   432       b = startn;
   433       pol.reset();
   434     }
   435 
   436     /// \brief Does one iteration
   437     ///
   438     /// If the Policy allows it searches for the best next node, then steps
   439     /// onto it.
   440     /// \return %False if one Policy condition wants to stop the search.
   441     bool  step()
   442     {
   443       ///Request premmision from ControllPolicy
   444       if( !pol.onStep() )
   445       return false;
   446 	
   447       ///Find the best next potential node
   448       Node n; bool found = false;
   449       for( EdgeIt e(gr,tabu[0]); e != INVALID; ++e )
   450       {
   451         Node m = (gr.source(e) == tabu[0]) ? gr.target(e) : gr.source(e);
   452         bool wrong = false;
   453         for( int i = 1; i != (signed int)tabu.size(); ++i )
   454           if( m == tabu[i] ) {
   455             wrong = true;
   456             break;
   457           }
   458         if( wrong )
   459           continue;
   460 
   461         if( !found ) {
   462           n = m;
   463           found = true;
   464         } else
   465           if( better(height[m], height[n]) ) {
   466             n = m;
   467           }
   468       }
   469 
   470       ///Handle stuck search
   471       if( !found ) {
   472         return pol.onStick();
   473       }
   474 
   475       ///Move on...
   476       tabu.push_front(n);
   477       while( mts < tabu.size() ) {
   478         tabu.pop_back();
   479       }
   480       if( better(height[n], height[b]) ) {
   481         b = n;
   482         if( !pol.onImprove(height[b]) )
   483         return false;
   484       }
   485 
   486       return true;
   487     }
   488 
   489     /// \brief Runs a search while the Policy stops it.
   490     ///
   491     /// \param startn The start node where the search begins.
   492     inline void  run( const Node startn ) {
   493       std::cin.unsetf( std::ios_base::skipws );
   494       char c;
   495       init( startn );
   496       while( step() )
   497       std::cin >> c;
   498       std::cin.setf( std::ios_base::skipws );
   499     }
   500 
   501     ///@}
   502 
   503     /// \name Query Functions
   504     /// The result of the TabuSearch algorithm can be obtained using these
   505     /// functions.\n
   506     ///@{
   507 
   508     /// \brief The node, the search is standing on.
   509     inline Node  current() const {
   510       return tabu[0];
   511     }
   512 
   513     /// \brief The best node found until now.
   514     inline Node  best() const {
   515       return b;
   516     }
   517 
   518     /// \brief Beginning to iterate on the current tabu set.
   519     inline TabuIterator  tabu_begin() const {
   520       return tabu.begin();
   521     }
   522 
   523     /// \brief Ending to iterate on the current tabu set.
   524     inline TabuIterator  tabu_end() const {
   525       return tabu.end();
   526     }
   527 
   528     ///@}
   529   };
   530 }
   531 #endif