lemon/simann.h
author alpar
Tue, 28 Nov 2006 17:15:50 +0000
changeset 2310 96cca167430a
parent 2242 16523135943d
child 2370 ed6539025f27
permissions -rw-r--r--
- update graph_oriantation.cc to be in line with the renamed xy and with the
new graphToEps() scaling.
- Fix and external directory build problem in Doxyfile.in
     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_SIMANN_H
    20 #define LEMON_SIMANN_H
    21 
    22 /// \ingroup experimental
    23 /// \file
    24 /// \brief Simulated annealing framework.
    25 ///
    26 /// \todo A test and some demo should be added
    27 /// \todo Doc should be improved
    28 /// \author Akos Ladanyi
    29 
    30 #include <cstdlib>
    31 #include <cmath>
    32 #include <limits>
    33 #include <lemon/time_measure.h>
    34 #include <lemon/random.h>
    35 
    36 namespace lemon {
    37 
    38 /// \addtogroup experimental
    39 /// @{
    40 
    41   class SimAnnBase;
    42 
    43   /// \brief A base class for controllers.
    44   class ControllerBase {
    45   public:
    46     friend class SimAnnBase;
    47     /// \brief Pointer to the simulated annealing base class.
    48     SimAnnBase *simann;
    49     /// \brief Initializes the controller.
    50     virtual void init() {}
    51     /// \brief This is called by the simulated annealing class when a
    52     /// neighbouring state gets accepted.
    53     virtual void acceptEvent() {}
    54     /// \brief This is called by the simulated annealing class when the
    55     /// accepted neighbouring state's cost is less than the best found one's.
    56     virtual void improveEvent() {}
    57     /// \brief This is called by the simulated annealing class when a
    58     /// neighbouring state gets rejected.
    59     virtual void rejectEvent() {}
    60     /// \brief Decides whether to continue the annealing process or not.
    61     virtual bool next() = 0;
    62     /// \brief Decides whether to accept the current solution or not.
    63     virtual bool accept() = 0;
    64     /// \brief Destructor.
    65     virtual ~ControllerBase() {}
    66   };
    67 
    68   /// \brief Skeleton of an entity class.
    69   class EntityBase {
    70   public:
    71     /// \brief Makes a minor change to the entity.
    72     /// \return the new cost
    73     virtual double mutate() = 0;
    74     /// \brief Restores the entity to its previous state i.e. reverts the
    75     /// effects of the last mutate().
    76     virtual void revert() = 0;
    77     /// \brief Makes a copy of the entity.
    78     virtual EntityBase* clone() = 0;
    79     /// \brief Makes a major change to the entity.
    80     virtual void randomize() = 0;
    81     /// \brief Destructor.
    82     virtual ~EntityBase() {}
    83   };
    84 
    85   /// \brief Simulated annealing abstract base class.
    86   ///
    87   /// Can be used to derive a custom simulated annealing class if \ref SimAnn
    88   /// doesn't fit your needs.
    89   class SimAnnBase {
    90   private:
    91     /// \brief Pointer to the controller.
    92     ControllerBase *controller;
    93     /// \brief Cost of the current solution.
    94     double curr_cost;
    95     /// \brief Cost of the best solution.
    96     double best_cost;
    97     /// \brief Cost of the previous solution.
    98     double prev_cost;
    99     /// \brief Cost of the solution preceding the previous one.
   100     double prev_prev_cost;
   101     /// \brief Number of iterations.
   102     long iter;
   103     /// \brief Number of iterations which did not improve the solution since
   104     /// the last improvement.
   105     long last_impr;
   106   protected:
   107     /// \brief Step to a neighbouring state.
   108     virtual double mutate() = 0;
   109     /// \brief Reverts the last mutate().
   110     virtual void revert() = 0;
   111     /// \brief Saves the current solution as the best one.
   112     virtual void saveAsBest() = 0;
   113     /// \brief Does initializations before each run.
   114     virtual void init() {
   115       controller->init();
   116       curr_cost = prev_cost = prev_prev_cost = best_cost =
   117         std::numeric_limits<double>::infinity();
   118       iter = last_impr = 0;
   119     }
   120   public:
   121     /// \brief Sets the controller class to use.
   122     void setController(ControllerBase &_controller) {
   123       controller = &_controller;
   124       controller->simann = this;
   125     }
   126     /// \brief Returns the cost of the current solution.
   127     double getCurrCost() const { return curr_cost; }
   128     /// \brief Returns the cost of the previous solution.
   129     double getPrevCost() const { return prev_cost; }
   130     /// \brief Returns the cost of the best solution.
   131     double getBestCost() const { return best_cost; }
   132     /// \brief Returns the number of iterations done.
   133     long getIter() const { return iter; }
   134     /// \brief Returns the ordinal number of the last iteration when the
   135     /// solution was improved.
   136     long getLastImpr() const { return last_impr; }
   137     /// \brief Performs one iteration.
   138     bool step() {
   139       iter++;
   140       prev_prev_cost = prev_cost;
   141       prev_cost = curr_cost;
   142       curr_cost = mutate();
   143       if (controller->accept()) {
   144         controller->acceptEvent();
   145         last_impr = iter;
   146         if (curr_cost < best_cost) {
   147           best_cost = curr_cost;
   148           saveAsBest();
   149           controller->improveEvent();
   150         }
   151       }
   152       else {
   153         revert();
   154         curr_cost = prev_cost;
   155         prev_cost = prev_prev_cost;
   156         controller->rejectEvent();
   157       }
   158       return controller->next();
   159     }
   160     /// \brief Performs a given number of iterations.
   161     /// \param n the number of iterations
   162     bool step(int n) {
   163       for(; n > 0 && step(); --n) ;
   164       return !n;
   165     }
   166     /// \brief Starts the annealing process.
   167     void run() {
   168       init();
   169       do { } while (step());
   170     }
   171     /// \brief Destructor.
   172     virtual ~SimAnnBase() {}
   173   };
   174 
   175   /// \brief Simulated annealing class.
   176   class SimAnn : public SimAnnBase {
   177   private:
   178     /// \brief Pointer to the current entity.
   179     EntityBase *curr_ent;
   180     /// \brief Pointer to the best entity.
   181     EntityBase *best_ent;
   182     /// \brief Does initializations before each run.
   183     void init() {
   184       SimAnnBase::init();
   185       if (best_ent) delete best_ent;
   186       best_ent = NULL;
   187       curr_ent->randomize();
   188     }
   189   public:
   190     /// \brief Constructor.
   191     SimAnn() : curr_ent(NULL), best_ent(NULL) {}
   192     /// \brief Destructor.
   193     virtual ~SimAnn() {
   194       if (best_ent) delete best_ent;
   195     }
   196     /// \brief Step to a neighbouring state.
   197     double mutate() {
   198       return curr_ent->mutate();
   199     }
   200     /// \brief Reverts the last mutate().
   201     void revert() {
   202       curr_ent->revert();
   203     }
   204     /// \brief Saves the current solution as the best one.
   205     void saveAsBest() { 
   206       if (best_ent) delete best_ent;
   207       best_ent = curr_ent->clone();
   208     }
   209     /// \brief Sets the current entity.
   210     void setEntity(EntityBase &_ent) {
   211       curr_ent = &_ent;
   212     }
   213     /// \brief Returns a copy of the best found entity.
   214     EntityBase* getBestEntity() { return best_ent->clone(); }
   215   };
   216 
   217   /// \brief A simple controller for the simulated annealing class.
   218   ///
   219   /// This controller starts from a given initial temperature and evenly
   220   /// decreases it.
   221   class SimpleController : public ControllerBase {
   222   private:
   223     /// \brief Maximum number of iterations.
   224     long max_iter;
   225     /// \brief Maximum number of iterations which do not improve the
   226     /// solution.
   227     long max_no_impr;
   228     /// \brief Temperature.
   229     double temp;
   230     /// \brief Annealing factor.
   231     double ann_fact;
   232     /// \brief Constructor.
   233     /// \param _max_iter maximum number of iterations
   234     /// \param _max_no_impr maximum number of consecutive iterations which do
   235     ///        not yield a better solution
   236     /// \param _temp initial temperature
   237     /// \param _ann_fact annealing factor
   238   public:
   239     SimpleController(long _max_iter = 500000, long _max_no_impr = 20000,
   240     double _temp = 1000.0, double _ann_fact = 0.9999) : max_iter(_max_iter),
   241       max_no_impr(_max_no_impr), temp(_temp), ann_fact(_ann_fact)
   242     {
   243     }
   244     /// \brief This is called when a neighbouring state gets accepted.
   245     void acceptEvent() {}
   246     /// \brief This is called when the accepted neighbouring state's cost is
   247     /// less than the best found one's.
   248     void improveEvent() {}
   249     /// \brief This is called when a neighbouring state gets rejected.
   250     void rejectEvent() {}
   251     /// \brief Decides whether to continue the annealing process or not. Also
   252     /// decreases the temperature.
   253     bool next() {
   254       temp *= ann_fact;
   255       bool quit = (simann->getIter() > max_iter) ||
   256         (simann->getIter() - simann->getLastImpr() > max_no_impr);
   257       return !quit;
   258     }
   259     /// \brief Decides whether to accept the current solution or not.
   260     bool accept() {
   261       double cost_diff = simann->getCurrCost() - simann->getPrevCost();
   262       return (rnd() <= exp(-(cost_diff / temp)));
   263     }
   264     /// \brief Destructor.
   265     virtual ~SimpleController() {}
   266   };
   267 
   268   /// \brief A controller with preset running time for the simulated annealing
   269   /// class.
   270   ///
   271   /// With this controller you can set the running time of the annealing
   272   /// process in advance. It works the following way: the controller measures
   273   /// a kind of divergence. The divergence is the difference of the average
   274   /// cost of the recently found solutions the cost of the best found one. In
   275   /// case this divergence is greater than a given threshold, then we decrease
   276   /// the annealing factor, that is we cool the system faster. In case the
   277   /// divergence is lower than the threshold, then we increase the temperature.
   278   /// The threshold is a function of the elapsed time which reaches zero at the
   279   /// desired end time.
   280   class AdvancedController : public ControllerBase {
   281   private:
   282     /// \brief Timer class to measure the elapsed time.
   283     Timer timer;
   284     /// \brief Calculates the threshold value.
   285     /// \param time the elapsed time in seconds
   286     virtual double threshold(double time) {
   287       return (-1.0) * start_threshold / end_time * time + start_threshold;
   288     }
   289     /// \brief Parameter used to calculate the running average.
   290     double alpha;
   291     /// \brief Parameter used to decrease the annealing factor.
   292     double beta;
   293     /// \brief Parameter used to increase the temperature.
   294     double gamma;
   295     /// \brief The time at the end of the algorithm.
   296     double end_time;
   297     /// \brief The time at the start of the algorithm.
   298     double start_time;
   299     /// \brief Starting threshold.
   300     double start_threshold;
   301     /// \brief Average cost of recent solutions.
   302     double avg_cost;
   303     /// \brief Temperature.
   304     double temp;
   305     /// \brief Annealing factor.
   306     double ann_fact;
   307     /// \brief Initial annealing factor.
   308     double init_ann_fact;
   309     /// \brief True when the annealing process has been started.
   310     bool start;
   311   public:
   312     /// \brief Constructor.
   313     /// \param _end_time running time in seconds
   314     /// \param _alpha parameter used to calculate the running average
   315     /// \param _beta parameter used to decrease the annealing factor
   316     /// \param _gamma parameter used to increase the temperature
   317     /// \param _ann_fact initial annealing factor
   318     AdvancedController(double _end_time, double _alpha = 0.2,
   319     double _beta = 0.9, double _gamma = 1.6, double _ann_fact = 0.9999) :
   320     alpha(_alpha), beta(_beta), gamma(_gamma), end_time(_end_time),
   321     ann_fact(_ann_fact), init_ann_fact(_ann_fact), start(false)
   322     {
   323     }
   324     /// \brief Does initializations before each run.
   325     void init() {
   326       avg_cost = simann->getCurrCost();
   327     }
   328     /// \brief This is called when a neighbouring state gets accepted.
   329     void acceptEvent() {
   330       avg_cost = alpha * simann->getCurrCost() + (1.0 - alpha) * avg_cost;
   331       if (!start) {
   332         static int cnt = 0;
   333         cnt++;
   334         if (cnt >= 100) {
   335           // calculate starting threshold and starting temperature
   336           start_threshold = 5.0 * fabs(simann->getBestCost() - avg_cost);
   337           temp = 10000.0;
   338           start = true;
   339           timer.restart();
   340         }
   341       }
   342     }
   343     /// \brief Decides whether to continue the annealing process or not.
   344     bool next() {
   345       if (!start) {
   346         return true;
   347       }
   348       else {
   349         double elapsed_time = timer.realTime();
   350         if (fabs(avg_cost - simann->getBestCost()) > threshold(elapsed_time)) {
   351           // decrease the annealing factor
   352           ann_fact *= beta;
   353         }
   354         else {
   355           // increase the temperature
   356           temp *= gamma;
   357           // reset the annealing factor
   358           ann_fact = init_ann_fact;
   359         }
   360         temp *= ann_fact;
   361         return elapsed_time < end_time;
   362       }
   363     }
   364     /// \brief Decides whether to accept the current solution or not.
   365     bool accept() {
   366       if (!start) {
   367         return true;
   368       }
   369       else {
   370         double cost_diff = simann->getCurrCost() - simann->getPrevCost();
   371         return (rnd() <= exp(-(cost_diff / temp)));
   372       }
   373     }
   374     /// \brief Destructor.
   375     virtual ~AdvancedController() {}
   376   };
   377 
   378 /// @}
   379 
   380 }
   381 
   382 #endif