Free cookie consent management tool by TermsFeed Policy Generator

source: branches/OKBJavaConnector/ECJClient/src/ec/select/BestSelection.java @ 9598

Last change on this file since 9598 was 6152, checked in by bfarka, 14 years ago

added ecj and custom statistics to communicate with the okb services #1441

File size: 5.4 KB
RevLine 
[6152]1/*
2  Copyright 2006 by Sean Luke
3  Licensed under the Academic Free License version 3.0
4  See the file "LICENSE" for more information
5*/
6
7
8package ec.select;
9import ec.util.*;
10import ec.*;
11/*
12 * BestSelection.java
13 *
14 * Created: Thu Feb 10 18:52:09 2000
15 * By: Sean Luke
16 */
17
18/**
19 * Picks among the best <i>n</i> individuals in a population in
20 * direct proportion to their absolute
21 * fitnesses as returned by their fitness() methods relative to the
22 * fitnesses of the other "best" individuals in that <i>n</i>.  This is expensive to
23 * set up and bring down, so it's not appropriate for steady-state evolution.
24 * If you're not familiar with the relative advantages of
25 * selection methods and just want a good one,
26 * use TournamentSelection instead.   Not appropriate for
27 * multiobjective fitnesses.
28 *
29 * <p><b><font color=red>
30 * Note: Fitnesses must be non-negative.  0 is assumed to be the worst fitness.
31 * </font></b>
32 *
33 <p><b>Typical Number of Individuals Produced Per <tt>produce(...)</tt> call</b><br>
34 Always 1.
35
36 <p><b>Parameters</b><br>
37 <table>
38 <tr><td valign=top><i>base.</i><tt>pick-worst</tt><br>
39 <font size=-1> bool = <tt>true</tt> or <tt>false</tt> (default)</font></td>
40 <td valign=top>(should we pick from among the <i>worst n</i> individuals in the tournament instead of the <i>best n</i>?)</td></tr>
41 <tr><td valign=top><i>base.</i><tt>n</tt><br>
42 <font size=-1> int > 0 (default is 1)</font></td>
43 <td valign=top>(the number of best-individuals to select from)</td></tr>
44 </table>
45
46 <p><b>Default Base</b><br>
47 select.best
48
49 * @author Sean Luke
50 * @version 1.0
51 */
52
53public class BestSelection extends SelectionMethod
54    {
55    /** Default base */
56    public static final String P_BEST = "best";
57    public static final String P_N = "n";
58    public static final String P_PICKWORST = "pick-worst";
59    /** Sorted, normalized, totalized fitnesses for the population */
60    public float[] sortedFit;
61    /** Sorted population -- since I *have* to use an int-sized
62        individual (short gives me only 16K),
63        I might as well just have pointers to the
64        population itself.  :-( */
65    public int[] sortedPop;
66
67    /** Do we pick the worst instead of the best? */
68    public boolean pickWorst;
69
70    public int bestn;
71
72    public Parameter defaultBase()
73        {
74        return SelectDefaults.base().push(P_BEST);
75        }
76
77    // don't need clone etc.
78
79    public void setup(final EvolutionState state, final Parameter base)
80        {
81        super.setup(state,base);
82       
83        Parameter def = defaultBase();
84       
85        bestn =
86            state.parameters.getInt(base.push(P_N),def.push(P_N),1);
87        if (bestn == 0 )
88            state.output.fatal("n must be an integer greater than 0", base.push(P_N),def.push(P_N));
89       
90        pickWorst = state.parameters.getBoolean(base.push(P_PICKWORST),def.push(P_PICKWORST),false);
91        }
92
93    public void prepareToProduce(final EvolutionState s,
94        final int subpopulation,
95        final int thread)
96        {
97        // load sortedPop integers
98        final Individual[] i = s.population.subpops[subpopulation].individuals;
99
100        sortedPop = new int[i.length];
101        for(int x=0;x<sortedPop.length;x++) sortedPop[x] = x;
102
103        // sort sortedPop in increasing fitness order
104        QuickSort.qsort(sortedPop,
105            new SortComparatorL()
106                {
107                public boolean lt(long a, long b)
108                    {
109                    return ((Individual)(i[(int)b])).fitness.betterThan(
110                        ((Individual)(i[(int)a])).fitness);
111                    }
112
113                public boolean gt(long a, long b)
114                    {
115                    return ((Individual)(i[(int)a])).fitness.betterThan(
116                        ((Individual)(i[(int)b])).fitness);
117                    }
118                });
119
120        // load sortedFit
121        sortedFit = new float[Math.min(sortedPop.length,bestn)];
122        if (pickWorst)
123            for(int x=0;x<sortedFit.length;x++)
124                sortedFit[x] = ((Individual)(i[sortedPop[x]])).fitness.fitness();
125        else
126            for(int x=0;x<sortedFit.length;x++)
127                sortedFit[x] = ((Individual)(i[sortedPop[sortedPop.length-x-1]])).fitness.fitness();
128
129        for(int x=0;x<sortedFit.length;x++)
130            {
131            if (sortedFit[x] < 0) // uh oh
132                s.output.fatal("Discovered a negative fitness value.  BestSelection requires that all fitness values be non-negative(offending subpopulation #" + subpopulation + ")");
133            }
134
135
136        // organize the distributions.  All zeros in fitness is fine
137        RandomChoice.organizeDistribution(sortedFit, true);
138        }
139
140    public int produce(final int subpopulation,
141        final EvolutionState state,
142        final int thread)
143        {
144        // Pick and return an individual from the population
145        if (pickWorst)
146            return sortedPop[RandomChoice.pickFromDistribution(
147                    sortedFit,state.random[thread].nextFloat())];
148        else
149            return sortedPop[sortedPop.length - RandomChoice.pickFromDistribution(
150                    sortedFit,state.random[thread].nextFloat()) - 1];           
151        }
152   
153    public void finishProducing(final EvolutionState s,
154        final int subpopulation,
155        final int thread)
156        {
157        // release the distributions so we can quickly
158        // garbage-collect them if necessary
159        sortedFit = null;
160        sortedPop = null;
161        }   
162    }
Note: See TracBrowser for help on using the repository browser.