lemon/simann.h
author alpar
Tue, 24 Oct 2006 16:49:41 +0000
changeset 2259 da142c310d02
parent 2229 4dbb6dd2dd4b
child 2304 108d6db4f32a
permissions -rw-r--r--
Spellcheck
alpar@1956
     1
/* -*- C++ -*-
alpar@1956
     2
 *
alpar@1956
     3
 * This file is a part of LEMON, a generic C++ optimization library
alpar@1956
     4
 *
alpar@1956
     5
 * Copyright (C) 2003-2006
alpar@1956
     6
 * Egervary Jeno Kombinatorikus Optimalizalasi Kutatocsoport
alpar@1956
     7
 * (Egervary Research Group on Combinatorial Optimization, EGRES).
alpar@1956
     8
 *
alpar@1956
     9
 * Permission to use, modify and distribute this software is granted
alpar@1956
    10
 * provided that this copyright notice appears in all copies. For
alpar@1956
    11
 * precise terms see the accompanying LICENSE file.
alpar@1956
    12
 *
alpar@1956
    13
 * This software is provided "AS IS" with no warranty of any kind,
alpar@1956
    14
 * express or implied, and with no claim as to its suitability for any
alpar@1956
    15
 * purpose.
alpar@1956
    16
 *
alpar@1956
    17
 */
alpar@1956
    18
alpar@1633
    19
#ifndef LEMON_SIMANN_H
alpar@1633
    20
#define LEMON_SIMANN_H
alpar@1633
    21
alpar@1633
    22
/// \ingroup experimental
alpar@1633
    23
/// \file
alpar@1633
    24
/// \brief Simulated annealing framework.
alpar@1847
    25
///
alpar@1847
    26
/// \todo A test and some demo should be added
alpar@1847
    27
/// \todo Doc should be improved
alpar@1633
    28
/// \author Akos Ladanyi
alpar@1633
    29
alpar@1633
    30
#include <cstdlib>
alpar@1633
    31
#include <cmath>
ladanyi@1918
    32
#include <limits>
alpar@1633
    33
#include <lemon/time_measure.h>
deba@2229
    34
#include <lemon/random.h>
deba@2035
    35
alpar@1633
    36
namespace lemon {
alpar@1633
    37
alpar@1633
    38
/// \addtogroup experimental
alpar@1633
    39
/// @{
alpar@1633
    40
deba@1932
    41
  class SimAnnBase;
deba@1932
    42
ladanyi@1918
    43
  /// \brief A base class for controllers.
alpar@1633
    44
  class ControllerBase {
ladanyi@1918
    45
  public:
alpar@1633
    46
    friend class SimAnnBase;
ladanyi@1918
    47
    /// \brief Pointer to the simulated annealing base class.
alpar@1633
    48
    SimAnnBase *simann;
ladanyi@1918
    49
    /// \brief Initializes the controller.
alpar@1633
    50
    virtual void init() {}
ladanyi@1918
    51
    /// \brief This is called by the simulated annealing class when a
ladanyi@1918
    52
    /// neighbouring state gets accepted.
alpar@1633
    53
    virtual void acceptEvent() {}
ladanyi@1918
    54
    /// \brief This is called by the simulated annealing class when the
ladanyi@1918
    55
    /// accepted neighbouring state's cost is less than the best found one's.
alpar@1633
    56
    virtual void improveEvent() {}
ladanyi@1918
    57
    /// \brief This is called by the simulated annealing class when a
ladanyi@1918
    58
    /// neighbouring state gets rejected.
alpar@1633
    59
    virtual void rejectEvent() {}
ladanyi@1918
    60
    /// \brief Decides whether to continue the annealing process or not.
alpar@1633
    61
    virtual bool next() = 0;
ladanyi@1918
    62
    /// \brief Decides whether to accept the current solution or not.
alpar@1633
    63
    virtual bool accept() = 0;
ladanyi@1918
    64
    /// \brief Destructor.
ladanyi@1918
    65
    virtual ~ControllerBase() {}
alpar@1633
    66
  };
alpar@1633
    67
ladanyi@1918
    68
  /// \brief Skeleton of an entity class.
alpar@1633
    69
  class EntityBase {
alpar@1633
    70
  public:
ladanyi@1918
    71
    /// \brief Makes a minor change to the entity.
ladanyi@1918
    72
    /// \return the new cost
alpar@1633
    73
    virtual double mutate() = 0;
ladanyi@1918
    74
    /// \brief Restores the entity to its previous state i.e. reverts the
ladanyi@1918
    75
    /// effects of the last mutate().
alpar@1633
    76
    virtual void revert() = 0;
ladanyi@1918
    77
    /// \brief Makes a copy of the entity.
alpar@1633
    78
    virtual EntityBase* clone() = 0;
ladanyi@1918
    79
    /// \brief Makes a major change to the entity.
alpar@1633
    80
    virtual void randomize() = 0;
ladanyi@1918
    81
    /// \brief Destructor.
ladanyi@1918
    82
    virtual ~EntityBase() {}
alpar@1633
    83
  };
alpar@1633
    84
ladanyi@1918
    85
  /// \brief Simulated annealing abstract base class.
ladanyi@1918
    86
  /// Can be used to derive a custom simulated annealing class if \ref SimAnn
ladanyi@1918
    87
  /// doesn't fit your needs.
alpar@1633
    88
  class SimAnnBase {
alpar@1633
    89
  private:
ladanyi@1918
    90
    /// \brief Pointer to the controller.
alpar@1633
    91
    ControllerBase *controller;
ladanyi@1918
    92
    /// \brief Cost of the current solution.
alpar@1633
    93
    double curr_cost;
ladanyi@1918
    94
    /// \brief Cost of the best solution.
alpar@1633
    95
    double best_cost;
ladanyi@1918
    96
    /// \brief Cost of the previous solution.
alpar@1633
    97
    double prev_cost;
ladanyi@1918
    98
    /// \brief Cost of the solution preceding the previous one.
alpar@1633
    99
    double prev_prev_cost;
ladanyi@1918
   100
    /// \brief Number of iterations.
alpar@1633
   101
    long iter;
ladanyi@1918
   102
    /// \brief Number of iterations which did not improve the solution since
ladanyi@1918
   103
    /// the last improvement.
alpar@1633
   104
    long last_impr;
alpar@1633
   105
  protected:
ladanyi@1918
   106
    /// \brief Step to a neighbouring state.
alpar@1633
   107
    virtual double mutate() = 0;
ladanyi@1918
   108
    /// \brief Reverts the last mutate().
alpar@1633
   109
    virtual void revert() = 0;
ladanyi@1918
   110
    /// \brief Saves the current solution as the best one.
alpar@1633
   111
    virtual void saveAsBest() = 0;
ladanyi@1918
   112
    /// \brief Does initializations before each run.
alpar@1633
   113
    virtual void init() {
alpar@1633
   114
      controller->init();
alpar@1633
   115
      curr_cost = prev_cost = prev_prev_cost = best_cost =
alpar@1633
   116
        std::numeric_limits<double>::infinity();
alpar@1633
   117
      iter = last_impr = 0;
alpar@1633
   118
    }
alpar@1633
   119
  public:
ladanyi@1918
   120
    /// \brief Sets the controller class to use.
alpar@1633
   121
    void setController(ControllerBase &_controller) {
alpar@1633
   122
      controller = &_controller;
alpar@1633
   123
      controller->simann = this;
alpar@1633
   124
    }
ladanyi@1918
   125
    /// \brief Returns the cost of the current solution.
alpar@1633
   126
    double getCurrCost() const { return curr_cost; }
ladanyi@1918
   127
    /// \brief Returns the cost of the previous solution.
alpar@1633
   128
    double getPrevCost() const { return prev_cost; }
ladanyi@1918
   129
    /// \brief Returns the cost of the best solution.
alpar@1633
   130
    double getBestCost() const { return best_cost; }
ladanyi@1918
   131
    /// \brief Returns the number of iterations done.
alpar@1633
   132
    long getIter() const { return iter; }
ladanyi@1918
   133
    /// \brief Returns the ordinal number of the last iteration when the
ladanyi@1918
   134
    /// solution was improved.
alpar@1633
   135
    long getLastImpr() const { return last_impr; }
ladanyi@1918
   136
    /// \brief Performs one iteration.
alpar@1633
   137
    bool step() {
alpar@1633
   138
      iter++;
alpar@1633
   139
      prev_prev_cost = prev_cost;
alpar@1633
   140
      prev_cost = curr_cost;
alpar@1633
   141
      curr_cost = mutate();
alpar@1633
   142
      if (controller->accept()) {
alpar@1633
   143
        controller->acceptEvent();
alpar@1633
   144
        last_impr = iter;
alpar@1633
   145
        if (curr_cost < best_cost) {
alpar@1633
   146
          best_cost = curr_cost;
alpar@1633
   147
          saveAsBest();
alpar@1633
   148
          controller->improveEvent();
alpar@1633
   149
        }
alpar@1633
   150
      }
alpar@1633
   151
      else {
alpar@1633
   152
        revert();
alpar@1633
   153
        curr_cost = prev_cost;
alpar@1633
   154
        prev_cost = prev_prev_cost;
alpar@1633
   155
        controller->rejectEvent();
alpar@1633
   156
      }
alpar@1633
   157
      return controller->next();
alpar@1633
   158
    }
ladanyi@1918
   159
    /// \brief Performs a given number of iterations.
ladanyi@1918
   160
    /// \param n the number of iterations
alpar@1633
   161
    bool step(int n) {
alpar@1633
   162
      for(; n > 0 && step(); --n) ;
alpar@1633
   163
      return !n;
alpar@1633
   164
    }
ladanyi@1918
   165
    /// \brief Starts the annealing process.
alpar@1633
   166
    void run() {
alpar@1633
   167
      init();
alpar@1633
   168
      do { } while (step());
alpar@1633
   169
    }
ladanyi@1918
   170
    /// \brief Destructor.
ladanyi@1918
   171
    virtual ~SimAnnBase() {}
alpar@1633
   172
  };
alpar@1633
   173
ladanyi@1918
   174
  /// \brief Simulated annealing class.
alpar@1633
   175
  class SimAnn : public SimAnnBase {
alpar@1633
   176
  private:
ladanyi@1918
   177
    /// \brief Pointer to the current entity.
alpar@1633
   178
    EntityBase *curr_ent;
ladanyi@1918
   179
    /// \brief Pointer to the best entity.
alpar@1633
   180
    EntityBase *best_ent;
ladanyi@1918
   181
    /// \brief Does initializations before each run.
alpar@1633
   182
    void init() {
alpar@1633
   183
      SimAnnBase::init();
alpar@1633
   184
      if (best_ent) delete best_ent;
alpar@1633
   185
      best_ent = NULL;
alpar@1633
   186
      curr_ent->randomize();
alpar@1633
   187
    }
alpar@1633
   188
  public:
ladanyi@1918
   189
    /// \brief Constructor.
alpar@1633
   190
    SimAnn() : curr_ent(NULL), best_ent(NULL) {}
ladanyi@1918
   191
    /// \brief Destructor.
alpar@1633
   192
    virtual ~SimAnn() {
alpar@1633
   193
      if (best_ent) delete best_ent;
alpar@1633
   194
    }
ladanyi@1918
   195
    /// \brief Step to a neighbouring state.
alpar@1633
   196
    double mutate() {
alpar@1633
   197
      return curr_ent->mutate();
alpar@1633
   198
    }
ladanyi@1918
   199
    /// \brief Reverts the last mutate().
alpar@1633
   200
    void revert() {
alpar@1633
   201
      curr_ent->revert();
alpar@1633
   202
    }
ladanyi@1918
   203
    /// \brief Saves the current solution as the best one.
alpar@1633
   204
    void saveAsBest() { 
alpar@1633
   205
      if (best_ent) delete best_ent;
alpar@1633
   206
      best_ent = curr_ent->clone();
alpar@1633
   207
    }
ladanyi@1918
   208
    /// \brief Sets the current entity.
alpar@1633
   209
    void setEntity(EntityBase &_ent) {
alpar@1633
   210
      curr_ent = &_ent;
alpar@1633
   211
    }
ladanyi@1918
   212
    /// \brief Returns a copy of the best found entity.
alpar@1633
   213
    EntityBase* getBestEntity() { return best_ent->clone(); }
alpar@1633
   214
  };
alpar@1633
   215
ladanyi@1918
   216
  /// \brief A simple controller for the simulated annealing class.
ladanyi@1918
   217
  /// This controller starts from a given initial temperature and evenly
ladanyi@1918
   218
  /// decreases it.
alpar@1633
   219
  class SimpleController : public ControllerBase {
ladanyi@1918
   220
  private:
ladanyi@1918
   221
    /// \brief Maximum number of iterations.
ladanyi@1918
   222
    long max_iter;
ladanyi@1918
   223
    /// \brief Maximum number of iterations which do not improve the
ladanyi@1918
   224
    /// solution.
ladanyi@1918
   225
    long max_no_impr;
ladanyi@1918
   226
    /// \brief Temperature.
ladanyi@1918
   227
    double temp;
ladanyi@1918
   228
    /// \brief Annealing factor.
ladanyi@1918
   229
    double ann_fact;
ladanyi@1918
   230
    /// \brief Constructor.
ladanyi@1918
   231
    /// \param _max_iter maximum number of iterations
ladanyi@1918
   232
    /// \param _max_no_impr maximum number of consecutive iterations which do
ladanyi@1918
   233
    ///        not yield a better solution
ladanyi@1918
   234
    /// \param _temp initial temperature
ladanyi@1918
   235
    /// \param _ann_fact annealing factor
alpar@1633
   236
  public:
alpar@1633
   237
    SimpleController(long _max_iter = 500000, long _max_no_impr = 20000,
alpar@1633
   238
    double _temp = 1000.0, double _ann_fact = 0.9999) : max_iter(_max_iter),
alpar@1633
   239
      max_no_impr(_max_no_impr), temp(_temp), ann_fact(_ann_fact)
alpar@1633
   240
    {
alpar@1633
   241
    }
ladanyi@1918
   242
    /// \brief This is called when a neighbouring state gets accepted.
alpar@1633
   243
    void acceptEvent() {}
ladanyi@1918
   244
    /// \brief This is called when the accepted neighbouring state's cost is
ladanyi@1918
   245
    /// less than the best found one's.
alpar@1633
   246
    void improveEvent() {}
ladanyi@1918
   247
    /// \brief This is called when a neighbouring state gets rejected.
alpar@1633
   248
    void rejectEvent() {}
ladanyi@1918
   249
    /// \brief Decides whether to continue the annealing process or not. Also
ladanyi@1918
   250
    /// decreases the temperature.
alpar@1633
   251
    bool next() {
alpar@1633
   252
      temp *= ann_fact;
alpar@1633
   253
      bool quit = (simann->getIter() > max_iter) ||
alpar@1633
   254
        (simann->getIter() - simann->getLastImpr() > max_no_impr);
alpar@1633
   255
      return !quit;
alpar@1633
   256
    }
ladanyi@1918
   257
    /// \brief Decides whether to accept the current solution or not.
alpar@1633
   258
    bool accept() {
ladanyi@1918
   259
      double cost_diff = simann->getCurrCost() - simann->getPrevCost();
deba@2242
   260
      return (rnd() <= exp(-(cost_diff / temp)));
alpar@1633
   261
    }
ladanyi@1918
   262
    /// \brief Destructor.
ladanyi@1918
   263
    virtual ~SimpleController() {}
alpar@1633
   264
  };
alpar@1633
   265
ladanyi@1918
   266
  /// \brief A controller with preset running time for the simulated annealing
ladanyi@1918
   267
  /// class.
ladanyi@1918
   268
  /// With this controller you can set the running time of the annealing
ladanyi@1918
   269
  /// process in advance. It works the following way: the controller measures
ladanyi@1918
   270
  /// a kind of divergence. The divergence is the difference of the average
ladanyi@1918
   271
  /// cost of the recently found solutions the cost of the best found one. In
ladanyi@1918
   272
  /// case this divergence is greater than a given threshold, then we decrease
ladanyi@1918
   273
  /// the annealing factor, that is we cool the system faster. In case the
ladanyi@1918
   274
  /// divergence is lower than the threshold, then we increase the temperature.
ladanyi@1918
   275
  /// The threshold is a function of the elapsed time which reaches zero at the
ladanyi@1918
   276
  /// desired end time.
alpar@1633
   277
  class AdvancedController : public ControllerBase {
alpar@1633
   278
  private:
ladanyi@1918
   279
    /// \brief Timer class to measure the elapsed time.
alpar@1633
   280
    Timer timer;
ladanyi@1918
   281
    /// \brief Calculates the threshold value.
ladanyi@1918
   282
    /// \param time the elapsed time in seconds
alpar@1633
   283
    virtual double threshold(double time) {
alpar@1633
   284
      return (-1.0) * start_threshold / end_time * time + start_threshold;
alpar@1633
   285
    }
ladanyi@1918
   286
    /// \brief Parameter used to calculate the running average.
ladanyi@1918
   287
    double alpha;
ladanyi@1918
   288
    /// \brief Parameter used to decrease the annealing factor.
ladanyi@1918
   289
    double beta;
ladanyi@1918
   290
    /// \brief Parameter used to increase the temperature.
ladanyi@1918
   291
    double gamma;
ladanyi@1918
   292
    /// \brief The time at the end of the algorithm.
ladanyi@1918
   293
    double end_time;
ladanyi@1918
   294
    /// \brief The time at the start of the algorithm.
ladanyi@1918
   295
    double start_time;
ladanyi@1918
   296
    /// \brief Starting threshold.
ladanyi@1918
   297
    double start_threshold;
ladanyi@1918
   298
    /// \brief Average cost of recent solutions.
ladanyi@1918
   299
    double avg_cost;
ladanyi@1918
   300
    /// \brief Temperature.
ladanyi@1918
   301
    double temp;
ladanyi@1918
   302
    /// \brief Annealing factor.
ladanyi@1918
   303
    double ann_fact;
ladanyi@1918
   304
    /// \brief Initial annealing factor.
ladanyi@1918
   305
    double init_ann_fact;
ladanyi@1918
   306
    /// \brief True when the annealing process has been started.
ladanyi@1918
   307
    bool start;
alpar@1633
   308
  public:
ladanyi@1918
   309
    /// \brief Constructor.
ladanyi@1918
   310
    /// \param _end_time running time in seconds
ladanyi@1918
   311
    /// \param _alpha parameter used to calculate the running average
ladanyi@1918
   312
    /// \param _beta parameter used to decrease the annealing factor
ladanyi@1918
   313
    /// \param _gamma parameter used to increase the temperature
ladanyi@1918
   314
    /// \param _ann_fact initial annealing factor
alpar@1633
   315
    AdvancedController(double _end_time, double _alpha = 0.2,
alpar@1633
   316
    double _beta = 0.9, double _gamma = 1.6, double _ann_fact = 0.9999) :
alpar@1633
   317
    alpha(_alpha), beta(_beta), gamma(_gamma), end_time(_end_time),
ladanyi@1918
   318
    ann_fact(_ann_fact), init_ann_fact(_ann_fact), start(false)
alpar@1633
   319
    {
alpar@1633
   320
    }
ladanyi@1918
   321
    /// \brief Does initializations before each run.
alpar@1633
   322
    void init() {
alpar@1633
   323
      avg_cost = simann->getCurrCost();
alpar@1633
   324
    }
ladanyi@1918
   325
    /// \brief This is called when a neighbouring state gets accepted.
alpar@1633
   326
    void acceptEvent() {
alpar@1633
   327
      avg_cost = alpha * simann->getCurrCost() + (1.0 - alpha) * avg_cost;
ladanyi@1918
   328
      if (!start) {
alpar@1633
   329
        static int cnt = 0;
alpar@1633
   330
        cnt++;
alpar@1633
   331
        if (cnt >= 100) {
alpar@1633
   332
          // calculate starting threshold and starting temperature
alpar@1633
   333
          start_threshold = 5.0 * fabs(simann->getBestCost() - avg_cost);
alpar@1633
   334
          temp = 10000.0;
ladanyi@1918
   335
          start = true;
alpar@1847
   336
          timer.restart();
alpar@1633
   337
        }
alpar@1633
   338
      }
alpar@1633
   339
    }
ladanyi@1918
   340
    /// \brief Decides whether to continue the annealing process or not.
alpar@1633
   341
    bool next() {
ladanyi@1918
   342
      if (!start) {
alpar@1633
   343
        return true;
alpar@1633
   344
      }
alpar@1633
   345
      else {
ladanyi@1918
   346
        double elapsed_time = timer.realTime();
alpar@1633
   347
        if (fabs(avg_cost - simann->getBestCost()) > threshold(elapsed_time)) {
alpar@1633
   348
          // decrease the annealing factor
alpar@1633
   349
          ann_fact *= beta;
alpar@1633
   350
        }
alpar@1633
   351
        else {
alpar@1633
   352
          // increase the temperature
alpar@1633
   353
          temp *= gamma;
alpar@1633
   354
          // reset the annealing factor
alpar@1633
   355
          ann_fact = init_ann_fact;
alpar@1633
   356
        }
alpar@1633
   357
        temp *= ann_fact;
alpar@1633
   358
        return elapsed_time < end_time;
alpar@1633
   359
      }
alpar@1633
   360
    }
ladanyi@1918
   361
    /// \brief Decides whether to accept the current solution or not.
alpar@1633
   362
    bool accept() {
ladanyi@1918
   363
      if (!start) {
alpar@1633
   364
        return true;
alpar@1633
   365
      }
alpar@1633
   366
      else {
ladanyi@1918
   367
        double cost_diff = simann->getCurrCost() - simann->getPrevCost();
deba@2242
   368
        return (rnd() <= exp(-(cost_diff / temp)));
alpar@1633
   369
      }
alpar@1633
   370
    }
ladanyi@1918
   371
    /// \brief Destructor.
ladanyi@1918
   372
    virtual ~AdvancedController() {}
alpar@1633
   373
  };
alpar@1633
   374
alpar@1633
   375
/// @}
alpar@1633
   376
alpar@1633
   377
}
alpar@1633
   378
alpar@1633
   379
#endif