lemon/simann.h
author kpeter
Thu, 13 Sep 2007 22:05:32 +0000
changeset 2470 46818ce27a60
parent 2370 ed6539025f27
child 2553 bfced05fa852
permissions -rw-r--r--
Small bug fixes.
     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_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   class SimAnnBase;
    39 
    40   /// \brief A base class for controllers.
    41   class ControllerBase {
    42   public:
    43     friend class SimAnnBase;
    44     /// \brief Pointer to the simulated annealing base class.
    45     SimAnnBase *simann;
    46     /// \brief Initializes the controller.
    47     virtual void init() {}
    48     /// \brief This is called by the simulated annealing class when a
    49     /// neighbouring state gets accepted.
    50     virtual void acceptEvent() {}
    51     /// \brief This is called by the simulated annealing class when the
    52     /// accepted neighbouring state's cost is less than the best found one's.
    53     virtual void improveEvent() {}
    54     /// \brief This is called by the simulated annealing class when a
    55     /// neighbouring state gets rejected.
    56     virtual void rejectEvent() {}
    57     /// \brief Decides whether to continue the annealing process or not.
    58     virtual bool next() = 0;
    59     /// \brief Decides whether to accept the current solution or not.
    60     virtual bool accept() = 0;
    61     /// \brief Destructor.
    62     virtual ~ControllerBase() {}
    63   };
    64 
    65   /// \brief Skeleton of an entity class.
    66   class EntityBase {
    67   public:
    68     /// \brief Makes a minor change to the entity.
    69     /// \return the new cost
    70     virtual double mutate() = 0;
    71     /// \brief Restores the entity to its previous state i.e. reverts the
    72     /// effects of the last mutate().
    73     virtual void revert() = 0;
    74     /// \brief Makes a copy of the entity.
    75     virtual EntityBase* clone() = 0;
    76     /// \brief Makes a major change to the entity.
    77     virtual void randomize() = 0;
    78     /// \brief Destructor.
    79     virtual ~EntityBase() {}
    80   };
    81 
    82   /// \brief Simulated annealing abstract base class.
    83   ///
    84   /// Can be used to derive a custom simulated annealing class if \ref SimAnn
    85   /// doesn't fit your needs.
    86   class SimAnnBase {
    87   private:
    88     /// \brief Pointer to the controller.
    89     ControllerBase *controller;
    90     /// \brief Cost of the current solution.
    91     double curr_cost;
    92     /// \brief Cost of the best solution.
    93     double best_cost;
    94     /// \brief Cost of the previous solution.
    95     double prev_cost;
    96     /// \brief Cost of the solution preceding the previous one.
    97     double prev_prev_cost;
    98     /// \brief Number of iterations.
    99     long iter;
   100     /// \brief Number of iterations which did not improve the solution since
   101     /// the last improvement.
   102     long last_impr;
   103   protected:
   104     /// \brief Step to a neighbouring state.
   105     virtual double mutate() = 0;
   106     /// \brief Reverts the last mutate().
   107     virtual void revert() = 0;
   108     /// \brief Saves the current solution as the best one.
   109     virtual void saveAsBest() = 0;
   110     /// \brief Does initializations before each run.
   111     virtual void init() {
   112       controller->init();
   113       curr_cost = prev_cost = prev_prev_cost = best_cost =
   114         std::numeric_limits<double>::infinity();
   115       iter = last_impr = 0;
   116     }
   117   public:
   118     /// \brief Sets the controller class to use.
   119     void setController(ControllerBase &_controller) {
   120       controller = &_controller;
   121       controller->simann = this;
   122     }
   123     /// \brief Returns the cost of the current solution.
   124     double getCurrCost() const { return curr_cost; }
   125     /// \brief Returns the cost of the previous solution.
   126     double getPrevCost() const { return prev_cost; }
   127     /// \brief Returns the cost of the best solution.
   128     double getBestCost() const { return best_cost; }
   129     /// \brief Returns the number of iterations done.
   130     long getIter() const { return iter; }
   131     /// \brief Returns the ordinal number of the last iteration when the
   132     /// solution was improved.
   133     long getLastImpr() const { return last_impr; }
   134     /// \brief Performs one iteration.
   135     bool step() {
   136       iter++;
   137       prev_prev_cost = prev_cost;
   138       prev_cost = curr_cost;
   139       curr_cost = mutate();
   140       if (controller->accept()) {
   141         controller->acceptEvent();
   142         last_impr = iter;
   143         if (curr_cost < best_cost) {
   144           best_cost = curr_cost;
   145           saveAsBest();
   146           controller->improveEvent();
   147         }
   148       }
   149       else {
   150         revert();
   151         curr_cost = prev_cost;
   152         prev_cost = prev_prev_cost;
   153         controller->rejectEvent();
   154       }
   155       return controller->next();
   156     }
   157     /// \brief Performs a given number of iterations.
   158     /// \param n the number of iterations
   159     bool step(int n) {
   160       for(; n > 0 && step(); --n) ;
   161       return !n;
   162     }
   163     /// \brief Starts the annealing process.
   164     void run() {
   165       init();
   166       do { } while (step());
   167     }
   168     /// \brief Destructor.
   169     virtual ~SimAnnBase() {}
   170   };
   171 
   172   /// \ingroup metah
   173   ///
   174   /// \brief Simulated annealing class.
   175   class SimAnn : public SimAnnBase {
   176   private:
   177     /// \brief Pointer to the current entity.
   178     EntityBase *curr_ent;
   179     /// \brief Pointer to the best entity.
   180     EntityBase *best_ent;
   181     /// \brief Does initializations before each run.
   182     void init() {
   183       SimAnnBase::init();
   184       if (best_ent) delete best_ent;
   185       best_ent = NULL;
   186       curr_ent->randomize();
   187     }
   188   public:
   189     /// \brief Constructor.
   190     SimAnn() : curr_ent(NULL), best_ent(NULL) {}
   191     /// \brief Destructor.
   192     virtual ~SimAnn() {
   193       if (best_ent) delete best_ent;
   194     }
   195     /// \brief Step to a neighbouring state.
   196     double mutate() {
   197       return curr_ent->mutate();
   198     }
   199     /// \brief Reverts the last mutate().
   200     void revert() {
   201       curr_ent->revert();
   202     }
   203     /// \brief Saves the current solution as the best one.
   204     void saveAsBest() { 
   205       if (best_ent) delete best_ent;
   206       best_ent = curr_ent->clone();
   207     }
   208     /// \brief Sets the current entity.
   209     void setEntity(EntityBase &_ent) {
   210       curr_ent = &_ent;
   211     }
   212     /// \brief Returns a copy of the best found entity.
   213     EntityBase* getBestEntity() { return best_ent->clone(); }
   214   };
   215 
   216   /// \brief A simple controller for the simulated annealing class.
   217   ///
   218   /// This controller starts from a given initial temperature and evenly
   219   /// decreases it.
   220   class SimpleController : public ControllerBase {
   221   private:
   222     /// \brief Maximum number of iterations.
   223     long max_iter;
   224     /// \brief Maximum number of iterations which do not improve the
   225     /// solution.
   226     long max_no_impr;
   227     /// \brief Temperature.
   228     double temp;
   229     /// \brief Annealing factor.
   230     double ann_fact;
   231     /// \brief Constructor.
   232     /// \param _max_iter maximum number of iterations
   233     /// \param _max_no_impr maximum number of consecutive iterations which do
   234     ///        not yield a better solution
   235     /// \param _temp initial temperature
   236     /// \param _ann_fact annealing factor
   237   public:
   238     SimpleController(long _max_iter = 500000, long _max_no_impr = 20000,
   239     double _temp = 1000.0, double _ann_fact = 0.9999) : max_iter(_max_iter),
   240       max_no_impr(_max_no_impr), temp(_temp), ann_fact(_ann_fact)
   241     {
   242     }
   243     /// \brief This is called when a neighbouring state gets accepted.
   244     void acceptEvent() {}
   245     /// \brief This is called when the accepted neighbouring state's cost is
   246     /// less than the best found one's.
   247     void improveEvent() {}
   248     /// \brief This is called when a neighbouring state gets rejected.
   249     void rejectEvent() {}
   250     /// \brief Decides whether to continue the annealing process or not. Also
   251     /// decreases the temperature.
   252     bool next() {
   253       temp *= ann_fact;
   254       bool quit = (simann->getIter() > max_iter) ||
   255         (simann->getIter() - simann->getLastImpr() > max_no_impr);
   256       return !quit;
   257     }
   258     /// \brief Decides whether to accept the current solution or not.
   259     bool accept() {
   260       double cost_diff = simann->getCurrCost() - simann->getPrevCost();
   261       return (rnd() <= exp(-(cost_diff / temp)));
   262     }
   263     /// \brief Destructor.
   264     virtual ~SimpleController() {}
   265   };
   266 
   267   /// \brief A controller with preset running time for the simulated annealing
   268   /// class.
   269   ///
   270   /// With this controller you can set the running time of the annealing
   271   /// process in advance. It works the following way: the controller measures
   272   /// a kind of divergence. The divergence is the difference of the average
   273   /// cost of the recently found solutions the cost of the best found one. In
   274   /// case this divergence is greater than a given threshold, then we decrease
   275   /// the annealing factor, that is we cool the system faster. In case the
   276   /// divergence is lower than the threshold, then we increase the temperature.
   277   /// The threshold is a function of the elapsed time which reaches zero at the
   278   /// desired end time.
   279   class AdvancedController : public ControllerBase {
   280   private:
   281     /// \brief Timer class to measure the elapsed time.
   282     Timer timer;
   283     /// \brief Calculates the threshold value.
   284     /// \param time the elapsed time in seconds
   285     virtual double threshold(double time) {
   286       return (-1.0) * start_threshold / end_time * time + start_threshold;
   287     }
   288     /// \brief Parameter used to calculate the running average.
   289     double alpha;
   290     /// \brief Parameter used to decrease the annealing factor.
   291     double beta;
   292     /// \brief Parameter used to increase the temperature.
   293     double gamma;
   294     /// \brief The time at the end of the algorithm.
   295     double end_time;
   296     /// \brief The time at the start of the algorithm.
   297     double start_time;
   298     /// \brief Starting threshold.
   299     double start_threshold;
   300     /// \brief Average cost of recent solutions.
   301     double avg_cost;
   302     /// \brief Temperature.
   303     double temp;
   304     /// \brief Annealing factor.
   305     double ann_fact;
   306     /// \brief Initial annealing factor.
   307     double init_ann_fact;
   308     /// \brief True when the annealing process has been started.
   309     bool start;
   310   public:
   311     /// \brief Constructor.
   312     /// \param _end_time running time in seconds
   313     /// \param _alpha parameter used to calculate the running average
   314     /// \param _beta parameter used to decrease the annealing factor
   315     /// \param _gamma parameter used to increase the temperature
   316     /// \param _ann_fact initial annealing factor
   317     AdvancedController(double _end_time, double _alpha = 0.2,
   318     double _beta = 0.9, double _gamma = 1.6, double _ann_fact = 0.9999) :
   319     alpha(_alpha), beta(_beta), gamma(_gamma), end_time(_end_time),
   320     ann_fact(_ann_fact), init_ann_fact(_ann_fact), start(false)
   321     {
   322     }
   323     /// \brief Does initializations before each run.
   324     void init() {
   325       avg_cost = simann->getCurrCost();
   326     }
   327     /// \brief This is called when a neighbouring state gets accepted.
   328     void acceptEvent() {
   329       avg_cost = alpha * simann->getCurrCost() + (1.0 - alpha) * avg_cost;
   330       if (!start) {
   331         static int cnt = 0;
   332         cnt++;
   333         if (cnt >= 100) {
   334           // calculate starting threshold and starting temperature
   335           start_threshold = 5.0 * fabs(simann->getBestCost() - avg_cost);
   336           temp = 10000.0;
   337           start = true;
   338           timer.restart();
   339         }
   340       }
   341     }
   342     /// \brief Decides whether to continue the annealing process or not.
   343     bool next() {
   344       if (!start) {
   345         return true;
   346       }
   347       else {
   348         double elapsed_time = timer.realTime();
   349         if (fabs(avg_cost - simann->getBestCost()) > threshold(elapsed_time)) {
   350           // decrease the annealing factor
   351           ann_fact *= beta;
   352         }
   353         else {
   354           // increase the temperature
   355           temp *= gamma;
   356           // reset the annealing factor
   357           ann_fact = init_ann_fact;
   358         }
   359         temp *= ann_fact;
   360         return elapsed_time < end_time;
   361       }
   362     }
   363     /// \brief Decides whether to accept the current solution or not.
   364     bool accept() {
   365       if (!start) {
   366         return true;
   367       }
   368       else {
   369         double cost_diff = simann->getCurrCost() - simann->getPrevCost();
   370         return (rnd() <= exp(-(cost_diff / temp)));
   371       }
   372     }
   373     /// \brief Destructor.
   374     virtual ~AdvancedController() {}
   375   };
   376 
   377 }
   378 
   379 #endif