nlpsolver/ThirdParty/EvolutionarySolver/src/net/adaptivebox/deps/behavior/PSGTBehavior.java

   1 /**
   2  * Description: The description of particle swarm (PS) Generate-and-test Behavior.
   3  *
   4   #Supported parameters:
   5   NAME    VALUE_type   Range      DefaultV        Description
   6   c1       real        [0, 2]     1.494           PSAgent: learning factor for pbest
   7   c2       real        [0, 2]     1.494           PSAgent: learning factor for gbest
   8   w        real        [0, 1]     0.729           PSAgent: inertia weight
   9   CL       real        [0, 0.1]   0               PSAgent: chaos factor
  10   //Other choices for c1, c2, w, and CL: (2, 2, 0.4, 0.001)
  11
  12  * Author          Create/Modi     Note
  13  * Xiaofeng Xie    May 11, 2004
  14  * Xiaofeng Xie    Jul 01, 2008
  15  *
  16  * This library is free software; you can redistribute it and/or
  17  * modify it under the terms of the GNU Lesser General Public
  18  * License as published by the Free Software Foundation; either
  19  * version 2.1 of the License, or (at your option) any later version.
  20  *
  21  * This library is distributed in the hope that it will be useful,
  22  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  23  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  24  * Lesser General Public License for more details.
  25  *
  26  * Please acknowledge the author(s) if you use this code in any way.
  27  *
  28  * @version 1.0
  29  * @Since MAOS1.0
  30  *
  31  * References:
  32  * [1] Kennedy J, Eberhart R C. Particle swarm optimization. IEEE Int. Conf. on
  33  * Neural Networks, Perth, Australia, 1995: 1942-1948
  34  *   For original particle swarm idea
  35  * [2] Shi Y H, Eberhart R C. A Modified Particle Swarm Optimizer. IEEE Inter. Conf.
  36  * on Evolutionary Computation, Anchorage, Alaska, 1998: 69-73
  37  *   For the inertia weight: adjust the trade-off between exploitation & exploration
  38  * [3] Clerc M, Kennedy J. The particle swarm - explosion, stability, and
  39  * convergence in a multidimensional complex space. IEEE Trans. on Evolutionary
  40  * Computation. 2002, 6 (1): 58-73
  41  *   Constriction factor: ensures the convergence
  42  * [4] Xie X F, Zhang W J, Yang Z L. A dissipative particle swarm optimization.
  43  * Congress on Evolutionary Computation, Hawaii, USA, 2002: 1456-1461
  44  *   The CL parameter
  45  * [5] Xie X F, Zhang W J, Bi D C. Optimizing semiconductor devices by self-
  46  * organizing particle swarm. Congress on Evolutionary Computation, Oregon, USA,
  47  * 2004: 2017-2022
  48  *   Further experimental analysis on the convergence of PSO
  49  * [6] X F Xie, W J Zhang. SWAF: swarm algorithm framework for numerical
  50  *     optimization. Genetic and Evolutionary Computation Conference (GECCO),
  51  *     Seattle, WA, USA, 2004: 238-250
  52  *     -> a generate-and-test behavior
  53  *
  54  */
  55
  56 package net.adaptivebox.deps.behavior;
  57
  58 import net.adaptivebox.global.RandomGenerator;
  59 import net.adaptivebox.goodness.IGoodnessCompareEngine;
  60 import net.adaptivebox.knowledge.Library;
  61 import net.adaptivebox.knowledge.SearchPoint;
  62 import net.adaptivebox.problem.ProblemEncoder;
  63 import net.adaptivebox.space.BasicPoint;
  64 import net.adaptivebox.space.DesignSpace;
  65
  66 public class PSGTBehavior extends AbsGTBehavior {
  67   // Two normally choices for (c1, c2, weight), i.e., (2, 2, 0.4), or (1.494,
  68   // 1.494, 0.729) The first is used in dissipative PSO (cf. [4]) as CL>0, and
  69   // the second is achieved by using constriction factors (cf. [3])
  70   public double c1 = 2;
  71   public double c2 = 2;
  72
  73   //inertia weight
  74   public double weight = 0.4;
  75
  76   //See ref[4], normally be 0.001~0.005
  77   public double CL = 0;
  78
  79   // the own memory: store the point that generated in old learning cycle
  80   private BasicPoint pold_t;
  81
  82   // the own memory: store the point that generated in last learning cycle
  83   private BasicPoint pcurrent_t;
  84
  85   @Override
  86   public void setMemPoints(SearchPoint pbest, BasicPoint pcurrent, BasicPoint pold) {
  87     pcurrent_t = pcurrent;
  88     pbest_t = pbest;
  89     pold_t = pold;
  90   }
  91
  92   @Override
  93   public void generateBehavior(SearchPoint trailPoint, ProblemEncoder problemEncoder) {
  94     DesignSpace designSpace = problemEncoder.getDesignSpace();
  95
  96     double[] pold_t_location = pold_t.getLocation();
  97     double[] pbest_t_location = pbest_t.getLocation();
  98     double[] pcurrent_t_location = pcurrent_t.getLocation();
  99     double[] gbest_t_location = socialLib.getGbest().getLocation();
 100     double[] trailPointLocation = trailPoint.getLocation();
 101
 102     int DIMENSION = designSpace.getDimension();
 103     for (int b = 0; b < DIMENSION; b++) {
 104       if (RandomGenerator.doubleZeroOneRandom() < CL) {
 105         designSpace.mutationAt(trailPointLocation, b);
 106         continue;
 107       }
 108
 109       double deltaxb = weight * (pcurrent_t_location[b] - pold_t_location[b])
 110             + c1 * RandomGenerator.doubleZeroOneRandom() * (pbest_t_location[b] - pcurrent_t_location[b])
 111             + c2 * RandomGenerator.doubleZeroOneRandom() * (gbest_t_location[b] - pcurrent_t_location[b]);
 112
 113       // limitation for delta_x
 114       double deltaxbm = 0.5 * designSpace.getMagnitudeIn(b);
 115
 116       if (deltaxb < -deltaxbm) {
 117         deltaxb = -deltaxbm;
 118       } else if (deltaxb > deltaxbm) {
 119         deltaxb = deltaxbm;
 120       }
 121
 122       trailPointLocation[b] = pcurrent_t_location[b] + deltaxb;
 123     }
 124   }
 125
 126   @Override
 127   public void testBehavior(SearchPoint trailPoint, IGoodnessCompareEngine qualityComparator) {
 128     Library.replace(qualityComparator, trailPoint, pbest_t);
 129     pold_t.importLocation(pcurrent_t);
 130     pcurrent_t.importLocation(trailPoint);
 131   }
 132 }