lemon/simann.h
author ladanyi
Tue, 23 Aug 2005 16:27:59 +0000
changeset 1647 d5dc956393b2
child 1847 7cbc12e42482
permissions -rw-r--r--
bugfix
     1 #ifndef LEMON_SIMANN_H
     2 #define LEMON_SIMANN_H
     3 
     4 /// \ingroup experimental
     5 /// \file
     6 /// \brief Simulated annealing framework.
     7 /// \author Akos Ladanyi
     8 
     9 #include <cstdlib>
    10 #include <cmath>
    11 #include <lemon/time_measure.h>
    12 
    13 namespace lemon {
    14 
    15 /// \addtogroup experimental
    16 /// @{
    17 
    18   /*! \brief A base class for controllers. */
    19   class ControllerBase {
    20     friend class SimAnnBase;
    21   public:
    22     /*! \brief Pointer to the simulated annealing base class. */
    23     SimAnnBase *simann;
    24     /*! \brief Initializes the controller. */
    25     virtual void init() {}
    26     /*! \brief This is called when a neighbouring state gets accepted. */
    27     virtual void acceptEvent() {}
    28     /*! \brief This is called when the accepted neighbouring state's cost is
    29      *  less than the best found one's.
    30      */
    31     virtual void improveEvent() {}
    32     /*! \brief This is called when a neighbouring state gets rejected. */
    33     virtual void rejectEvent() {}
    34     /*! \brief Decides whether to continue the annealing process or not. */
    35     virtual bool next() = 0;
    36     /*! \brief Decides whether to accept the current solution or not. */
    37     virtual bool accept() = 0;
    38   };
    39 
    40   /*! \brief Skeleton of an entity class. */
    41   class EntityBase {
    42   public:
    43     /*! \brief Makes a minor change to the entity.
    44      *  \return the new cost
    45      */
    46     virtual double mutate() = 0;
    47     /*! \brief Restores the entity to its previous state i.e. reverts the
    48      *  effects of the last mutate().
    49      */
    50     virtual void revert() = 0;
    51     /*! \brief Makes a copy of the entity. */
    52     virtual EntityBase* clone() = 0;
    53     /*! \brief Makes a major change to the entity. */
    54     virtual void randomize() = 0;
    55   };
    56 
    57   /*! \brief Simulated annealing base class. */
    58   class SimAnnBase {
    59   private:
    60     /*! Pointer to the controller. */
    61     ControllerBase *controller;
    62     /*! \brief Cost of the current solution. */
    63     double curr_cost;
    64     /*! \brief Cost of the best solution. */
    65     double best_cost;
    66     /*! \brief Cost of the previous solution. */
    67     double prev_cost;
    68     /*! \brief Cost of the solution preceding the previous one. */
    69     double prev_prev_cost;
    70     /*! \brief Number of iterations. */
    71     long iter;
    72     /*! \brief Number of iterations which did not improve the solution since
    73      *  the last improvement. */
    74     long last_impr;
    75   protected:
    76     /*! \brief Step to a neighbouring state. */
    77     virtual double mutate() = 0;
    78     /*! \brief Reverts the last mutate(). */
    79     virtual void revert() = 0;
    80     /*! \brief Saves the current solution as the best one. */
    81     virtual void saveAsBest() = 0;
    82     /*! \brief Does initializations before each run. */
    83     virtual void init() {
    84       controller->init();
    85       curr_cost = prev_cost = prev_prev_cost = best_cost =
    86         std::numeric_limits<double>::infinity();
    87       iter = last_impr = 0;
    88     }
    89   public:
    90     /*! \brief Sets the controller class to use. */
    91     void setController(ControllerBase &_controller) {
    92       controller = &_controller;
    93       controller->simann = this;
    94     }
    95     /*! \brief Returns the cost of the current solution. */
    96     double getCurrCost() const { return curr_cost; }
    97     /*! \brief Returns the cost of the previous solution. */
    98     double getPrevCost() const { return prev_cost; }
    99     /*! \brief Returns the cost of the best solution. */
   100     double getBestCost() const { return best_cost; }
   101     /*! \brief Returns the number of iterations. */
   102     long getIter() const { return iter; }
   103     /*! \brief Returns the number of the last iteration when the solution was
   104      *  improved.
   105      */
   106     long getLastImpr() const { return last_impr; }
   107     /*! \brief Performs one iteration. */
   108     bool step() {
   109       iter++;
   110       prev_prev_cost = prev_cost;
   111       prev_cost = curr_cost;
   112       curr_cost = mutate();
   113       if (controller->accept()) {
   114         controller->acceptEvent();
   115         last_impr = iter;
   116         if (curr_cost < best_cost) {
   117           best_cost = curr_cost;
   118           saveAsBest();
   119           controller->improveEvent();
   120         }
   121       }
   122       else {
   123         revert();
   124         curr_cost = prev_cost;
   125         prev_cost = prev_prev_cost;
   126         controller->rejectEvent();
   127       }
   128       return controller->next();
   129     }
   130     /*! \brief Performs a given number of iterations.
   131      *  \param n the number of iterations
   132      */
   133     bool step(int n) {
   134       for(; n > 0 && step(); --n) ;
   135       return !n;
   136     }
   137     /*! \brief Starts the annealing process. */
   138     void run() {
   139       init();
   140       do { } while (step());
   141     }
   142   };
   143 
   144   /*! \brief Simulated annealing class. */
   145   class SimAnn : public SimAnnBase {
   146   private:
   147     /*! \brief Pointer to the current entity. */
   148     EntityBase *curr_ent;
   149     /*! \brief Pointer to the best entity. */
   150     EntityBase *best_ent;
   151     /*! \brief Does initializations before each run. */
   152     void init() {
   153       SimAnnBase::init();
   154       if (best_ent) delete best_ent;
   155       best_ent = NULL;
   156       curr_ent->randomize();
   157     }
   158   public:
   159     /*! \brief Constructor. */
   160     SimAnn() : curr_ent(NULL), best_ent(NULL) {}
   161     /*! \brief Destructor. */
   162     virtual ~SimAnn() {
   163       if (best_ent) delete best_ent;
   164     }
   165     /*! \brief Step to a neighbouring state. */
   166     double mutate() {
   167       return curr_ent->mutate();
   168     }
   169     /*! \brief Reverts the last mutate(). */
   170     void revert() {
   171       curr_ent->revert();
   172     }
   173     /*! \brief Saves the current solution as the best one. */
   174     void saveAsBest() { 
   175       if (best_ent) delete best_ent;
   176       best_ent = curr_ent->clone();
   177     }
   178     /*! \brief Sets the current entity. */
   179     void setEntity(EntityBase &_ent) {
   180       curr_ent = &_ent;
   181     }
   182     /*! \brief Returns a copy of the best found entity. */
   183     EntityBase* getBestEntity() { return best_ent->clone(); }
   184   };
   185 
   186   /*! \brief A simple controller for the simulated annealing class. */
   187   class SimpleController : public ControllerBase {
   188   public:
   189     /*! \brief Maximum number of iterations. */
   190     long max_iter;
   191     /*! \brief Maximum number of iterations which do not improve the
   192      *  solution. */
   193     long max_no_impr;
   194     /*! \brief Temperature. */
   195     double temp;
   196     /*! \brief Annealing factor. */
   197     double ann_fact;
   198     /*! \brief Constructor.
   199      *  \param _max_iter maximum number of iterations
   200      *  \param _max_no_impr maximum number of consecutive iterations which do
   201      *         not yield a better solution
   202      *  \param _temp initial temperature
   203      *  \param _ann_fact annealing factor
   204      */
   205     SimpleController(long _max_iter = 500000, long _max_no_impr = 20000,
   206     double _temp = 1000.0, double _ann_fact = 0.9999) : max_iter(_max_iter),
   207       max_no_impr(_max_no_impr), temp(_temp), ann_fact(_ann_fact)
   208     {
   209       srand48(time(0));
   210     }
   211     /*! \brief This is called when a neighbouring state gets accepted. */
   212     void acceptEvent() {}
   213     /*! \brief This is called when the accepted neighbouring state's cost is
   214      *  less than the best found one's.
   215      */
   216     void improveEvent() {}
   217     /*! \brief This is called when a neighbouring state gets rejected. */
   218     void rejectEvent() {}
   219     /*! \brief Decides whether to continue the annealing process or not. Also
   220      *  decreases the temperature. */
   221     bool next() {
   222       temp *= ann_fact;
   223       bool quit = (simann->getIter() > max_iter) ||
   224         (simann->getIter() - simann->getLastImpr() > max_no_impr);
   225       return !quit;
   226     }
   227     /*! \brief Decides whether to accept the current solution or not. */
   228     bool accept() {
   229       double cost_diff = simann->getPrevCost() - simann->getCurrCost();
   230       return (drand48() <= exp(cost_diff / temp));
   231     }
   232   };
   233 
   234   /*! \brief A controller with preset running time for the simulated annealing
   235    *  class.
   236    *
   237    *  With this controller you can set the running time of the annealing
   238    *  process in advance. It works the following way: the controller measures
   239    *  a kind of divergence. The divergence is the difference of the average
   240    *  cost of the recently found solutions the cost of the best found one. In
   241    *  case this divergence is greater than a given threshold, then we decrease
   242    *  the annealing factor, that is we cool the system faster. In case the
   243    *  divergence is lower than the threshold, then we increase the temperature.
   244    *  The threshold is a function of the elapsed time which reaches zero at the
   245    *  desired end time.
   246    */
   247   class AdvancedController : public ControllerBase {
   248   private:
   249     Timer timer;
   250     /*! \param time the elapsed time in seconds */
   251     virtual double threshold(double time) {
   252       return (-1.0) * start_threshold / end_time * time + start_threshold;
   253     }
   254   public:
   255     double alpha;
   256     double beta;
   257     double gamma;
   258     /*! \brief The time at the end of the algorithm. */
   259     double end_time;
   260     /*! \brief The time at the start of the algorithm. */
   261     double start_time;
   262     /*! \brief Starting threshold. */
   263     double start_threshold;
   264     /*! \brief Average cost of recent solutions. */
   265     double avg_cost;
   266     /*! \brief Temperature. */
   267     double temp;
   268     /*! \brief Annealing factor. */
   269     double ann_fact;
   270     /*! \brief Initial annealing factor. */
   271     double init_ann_fact;
   272     bool warmup;
   273     /*! \brief Constructor.
   274      *  \param _end_time running time in seconds
   275      *  \param _alpha parameter used to calculate the running average
   276      *  \param _beta parameter used to decrease the annealing factor
   277      *  \param _gamma parameter used to increase the temperature
   278      *  \param _ann_fact initial annealing factor
   279      */
   280     AdvancedController(double _end_time, double _alpha = 0.2,
   281     double _beta = 0.9, double _gamma = 1.6, double _ann_fact = 0.9999) :
   282     alpha(_alpha), beta(_beta), gamma(_gamma), end_time(_end_time),
   283     ann_fact(_ann_fact), init_ann_fact(_ann_fact), warmup(true)
   284     {
   285       srand48(time(0));
   286     }
   287     void init() {
   288       avg_cost = simann->getCurrCost();
   289     }
   290     /*! \brief This is called when a neighbouring state gets accepted. */
   291     void acceptEvent() {
   292       avg_cost = alpha * simann->getCurrCost() + (1.0 - alpha) * avg_cost;
   293       if (warmup) {
   294         static int cnt = 0;
   295         cnt++;
   296         if (cnt >= 100) {
   297           // calculate starting threshold and starting temperature
   298           start_threshold = 5.0 * fabs(simann->getBestCost() - avg_cost);
   299           temp = 10000.0;
   300           warmup = false;
   301           timer.reset();
   302         }
   303       }
   304     }
   305     /*! \brief Decides whether to continue the annealing process or not. */
   306     bool next() {
   307       if (warmup) {
   308         return true;
   309       }
   310       else {
   311         double elapsed_time = timer.getRealTime();
   312         if (fabs(avg_cost - simann->getBestCost()) > threshold(elapsed_time)) {
   313           // decrease the annealing factor
   314           ann_fact *= beta;
   315         }
   316         else {
   317           // increase the temperature
   318           temp *= gamma;
   319           // reset the annealing factor
   320           ann_fact = init_ann_fact;
   321         }
   322         temp *= ann_fact;
   323         return elapsed_time < end_time;
   324       }
   325     }
   326     /*! \brief Decides whether to accept the current solution or not. */
   327     bool accept() {
   328       if (warmup) {
   329         // we accept eveything during the "warm up" phase
   330         return true;
   331       }
   332       else {
   333         double cost_diff = simann->getPrevCost() - simann->getCurrCost();
   334         if (cost_diff < 0.0) {
   335           return (drand48() <= exp(cost_diff / temp));
   336         }
   337         else {
   338           return true;
   339         }
   340       }
   341     }
   342   };
   343 
   344 /// @}
   345 
   346 }
   347 
   348 #endif