[6152] | 1 | /* |
---|
| 2 | Copyright 2006 by Sean Luke and George Mason University |
---|
| 3 | Licensed under the Academic Free License version 3.0 |
---|
| 4 | See the file "LICENSE" for more information |
---|
| 5 | */ |
---|
| 6 | |
---|
| 7 | |
---|
| 8 | package ec.parsimony; |
---|
| 9 | |
---|
| 10 | import ec.*; |
---|
| 11 | import ec.util.*; |
---|
| 12 | import ec.steadystate.*; |
---|
| 13 | import ec.select.*; |
---|
| 14 | |
---|
| 15 | /* |
---|
| 16 | * BucketTournamentSelection.java |
---|
| 17 | * |
---|
| 18 | * Created: Mon Apr 09 17:02:30 2001 |
---|
| 19 | * By: Liviu Panait |
---|
| 20 | */ |
---|
| 21 | |
---|
| 22 | /** |
---|
| 23 | * |
---|
| 24 | * Does a tournament selection, limited to the subpopulation it's |
---|
| 25 | * working in at the time. |
---|
| 26 | * |
---|
| 27 | * <p>Bucket Lexicographic Tournament selection works like as follows. There is a |
---|
| 28 | * number of buckets (<i>num-buckets</i>) specified beforehand, and each is |
---|
| 29 | * assigned a rank from 1 to <i>num-buckets</i>. The population, of size <i>pop-size</i>, |
---|
| 30 | * is sorted by fitness. The bottom <i>pop-size</i>/<i>num-buckets</i> individuals are |
---|
| 31 | * placed in the worst ranked bucket, plus any individuals remaining in the population with |
---|
| 32 | * the same fitness as the best individual in the bucket. Then the second worst |
---|
| 33 | * <i>pop-size</i>/<i>num-buckets</i> individuals are placed in the second worst ranked bucket, |
---|
| 34 | * plus any individuals in the population equal in fitness to the best individual in that bucket. |
---|
| 35 | * This continues until there are no individuals in the population. Note that the topmost bucket |
---|
| 36 | * with individuals can hold fewer than <i>pop-size</i>/<i>num-buckets</i> individuals, if |
---|
| 37 | * <i>pop-size</i> is not a multiple of <i>num-buckets</i>. Depending on the number of |
---|
| 38 | * equal-fitness individuals in the population, there can be some top buckets that are never |
---|
| 39 | * filled. The fitness of each individual in a bucket is set to the rank of the bucket holding |
---|
| 40 | * it. Direct bucketing has the effect of trading off fitness differences for size. Thus the |
---|
| 41 | * larger the bucket, the stronger the emphasis on size as a secondary objective. |
---|
| 42 | * |
---|
| 43 | * After ranking the individuals, <i>size</i> individuals are chosen at random from the |
---|
| 44 | * population. Of those individuals, the one with the highest rank is selected. If the two |
---|
| 45 | * individuals are in the same rank, meaning that they have similar fitness, the one |
---|
| 46 | * with the smallest size is selected. |
---|
| 47 | * |
---|
| 48 | * <p>Bucket Lexicographic Tournament selection is so simple that it doesn't |
---|
| 49 | * need to maintain a cache of any form, so many of the SelectionMethod methods |
---|
| 50 | * just don't do anything at all. |
---|
| 51 | * |
---|
| 52 | |
---|
| 53 | <p><b>Typical Number of Individuals Produced Per <tt>produce(...)</tt> call</b><br> |
---|
| 54 | Always 1. |
---|
| 55 | |
---|
| 56 | <p><b>Parameters</b><br> |
---|
| 57 | <table> |
---|
| 58 | <tr><td valign=top><i>base.</i><tt>size</tt><br> |
---|
| 59 | <font size=-1>int >= 1 (default 7)</font></td> |
---|
| 60 | <td valign=top>(the tournament size)</td></tr> |
---|
| 61 | |
---|
| 62 | <tr><td valign=top><i>base.</i><tt>pick-worst</tt><br> |
---|
| 63 | <font size=-1> bool = <tt>true</tt> or <tt>false</tt> (default)</font></td> |
---|
| 64 | <td valign=top>(should we pick the <i>worst</i> individual in the tournament instead of the <i>best</i>?)</td></tr> |
---|
| 65 | |
---|
| 66 | <tr><td valign=top><i>base.</i><tt>num-buckets</tt><br> |
---|
| 67 | <font size=-1>int >= 1 (default 10)</font></td> |
---|
| 68 | <td valign=top>(the number of buckets)</td></tr> |
---|
| 69 | </table> |
---|
| 70 | |
---|
| 71 | <p><b>Default Base</b><br> |
---|
| 72 | select.bucket-tournament |
---|
| 73 | |
---|
| 74 | * |
---|
| 75 | * @author Liviu Panait |
---|
| 76 | * @version 1.0 |
---|
| 77 | */ |
---|
| 78 | |
---|
| 79 | public class BucketTournamentSelection extends SelectionMethod implements SteadyStateBSourceForm |
---|
| 80 | { |
---|
| 81 | /** Default base */ |
---|
| 82 | public static final String P_TOURNAMENT = "bucket-tournament"; |
---|
| 83 | |
---|
| 84 | /** If the worst individual should be picked in the tournament */ |
---|
| 85 | public static final String P_PICKWORST = "pick-worst"; |
---|
| 86 | |
---|
| 87 | /** Tournament size parameter */ |
---|
| 88 | public static final String P_SIZE = "size"; |
---|
| 89 | |
---|
| 90 | /** Default size */ |
---|
| 91 | public static final int DEFAULT_SIZE = 7; |
---|
| 92 | |
---|
| 93 | /** The number of buckets */ |
---|
| 94 | public static final String P_BUCKETS = "num-buckets"; |
---|
| 95 | |
---|
| 96 | /** Default number of buckets */ |
---|
| 97 | public static final int N_BUCKETS_DEFAULT = 10; |
---|
| 98 | |
---|
| 99 | /** Size of the tournament*/ |
---|
| 100 | public int size; |
---|
| 101 | |
---|
| 102 | /** Do we pick the worst instead of the best? */ |
---|
| 103 | public boolean pickWorst; |
---|
| 104 | |
---|
| 105 | // the number of buckets |
---|
| 106 | int nBuckets; |
---|
| 107 | |
---|
| 108 | // the indexes of the buckets where the individuals should go (will be used instead of fitness) |
---|
| 109 | int[] bucketValues; |
---|
| 110 | |
---|
| 111 | public Parameter defaultBase() |
---|
| 112 | { |
---|
| 113 | return SelectDefaults.base().push(P_TOURNAMENT); |
---|
| 114 | } |
---|
| 115 | |
---|
| 116 | public void setup(final EvolutionState state, final Parameter base) |
---|
| 117 | { |
---|
| 118 | super.setup(state,base); |
---|
| 119 | |
---|
| 120 | Parameter def = defaultBase(); |
---|
| 121 | |
---|
| 122 | size = state.parameters.getInt(base.push(P_SIZE),def.push(P_SIZE),1); |
---|
| 123 | if (size < 1) |
---|
| 124 | state.output.fatal("Tournament size must be >= 1.",base.push(P_SIZE),def.push(P_SIZE)); |
---|
| 125 | |
---|
| 126 | if( state.parameters.exists( base.push(P_BUCKETS), def.push(P_BUCKETS))) |
---|
| 127 | { |
---|
| 128 | nBuckets = state.parameters.getInt(base.push(P_BUCKETS),def.push(P_BUCKETS),1); |
---|
| 129 | if (nBuckets < 1) |
---|
| 130 | { |
---|
| 131 | state.output.fatal("The number of buckets size must be >= 1.",base.push(P_BUCKETS),def.push(P_BUCKETS)); |
---|
| 132 | } |
---|
| 133 | } |
---|
| 134 | else |
---|
| 135 | { |
---|
| 136 | nBuckets = N_BUCKETS_DEFAULT; |
---|
| 137 | } |
---|
| 138 | |
---|
| 139 | pickWorst = state.parameters.getBoolean(base.push(P_PICKWORST),def.push(P_PICKWORST),false); |
---|
| 140 | } |
---|
| 141 | |
---|
| 142 | /** Prepare to produce: create the buckets!!!! */ |
---|
| 143 | public void prepareToProduce(final EvolutionState state, final int subpopulation, final int thread) |
---|
| 144 | { |
---|
| 145 | bucketValues = new int[ state.population.subpops[subpopulation].individuals.length ]; |
---|
| 146 | |
---|
| 147 | // correct? |
---|
| 148 | java.util.Arrays.sort(state.population.subpops[subpopulation].individuals, |
---|
| 149 | new java.util.Comparator() |
---|
| 150 | { |
---|
| 151 | public int compare(Object o1, Object o2) |
---|
| 152 | { |
---|
| 153 | Individual a = (Individual) o1; |
---|
| 154 | Individual b = (Individual) o2; |
---|
| 155 | if (a.fitness.betterThan(b.fitness)) |
---|
| 156 | return 1; |
---|
| 157 | if (b.fitness.betterThan(a.fitness)) |
---|
| 158 | return -1; |
---|
| 159 | return 0; |
---|
| 160 | } |
---|
| 161 | }); |
---|
| 162 | |
---|
| 163 | |
---|
| 164 | // how many individuals in current bucket |
---|
| 165 | int nInd; |
---|
| 166 | |
---|
| 167 | float averageBuck = ((float)state.population.subpops[subpopulation].individuals.length)/ |
---|
| 168 | ((float)nBuckets); |
---|
| 169 | |
---|
| 170 | // first individual goes into first bucket |
---|
| 171 | bucketValues[0] = 0; |
---|
| 172 | |
---|
| 173 | // now there is one individual in the first bucket |
---|
| 174 | nInd = 1; |
---|
| 175 | |
---|
| 176 | for( int i = 1 ; i < state.population.subpops[subpopulation].individuals.length ; i++ ) |
---|
| 177 | { |
---|
| 178 | // if there is still some place left in the current bucket, throw the current individual there too |
---|
| 179 | if( nInd < averageBuck ) |
---|
| 180 | { |
---|
| 181 | bucketValues[i] = bucketValues[i-1]; |
---|
| 182 | nInd++; |
---|
| 183 | } |
---|
| 184 | else // check if it has the same fitness as last individual |
---|
| 185 | { |
---|
| 186 | if( ((Individual)state.population.subpops[subpopulation].individuals[i]).fitness.equivalentTo( |
---|
| 187 | ((Individual)state.population.subpops[subpopulation].individuals[i-1]).fitness ) ) |
---|
| 188 | { |
---|
| 189 | // now the individual has exactly the same fitness as previous one, |
---|
| 190 | // so we just put it in the same bucket as the previous one(s) |
---|
| 191 | bucketValues[i] = bucketValues[i-1]; |
---|
| 192 | nInd++; |
---|
| 193 | } |
---|
| 194 | else |
---|
| 195 | { |
---|
| 196 | // if there are buckets left |
---|
| 197 | if( bucketValues[i-1]+1 < nBuckets ) |
---|
| 198 | { |
---|
| 199 | // new bucket!!!! |
---|
| 200 | bucketValues[i] = bucketValues[i-1] - 1; |
---|
| 201 | // with only one individual |
---|
| 202 | nInd = 1; |
---|
| 203 | } |
---|
| 204 | else // no more buckets left, just stick everything in the last bucket |
---|
| 205 | { |
---|
| 206 | bucketValues[i] = bucketValues[i-1]; |
---|
| 207 | nInd++; |
---|
| 208 | } |
---|
| 209 | } |
---|
| 210 | } |
---|
| 211 | } |
---|
| 212 | } |
---|
| 213 | |
---|
| 214 | public int produce(final int subpopulation, |
---|
| 215 | final EvolutionState state, |
---|
| 216 | final int thread) |
---|
| 217 | { |
---|
| 218 | // pick size random individuals, then pick the best. |
---|
| 219 | Individual[] oldinds = (state.population.subpops[subpopulation].individuals); |
---|
| 220 | int i = state.random[thread].nextInt(oldinds.length); |
---|
| 221 | long si = 0; |
---|
| 222 | |
---|
| 223 | for (int x=1;x<size;x++) |
---|
| 224 | { |
---|
| 225 | int j = state.random[thread].nextInt(oldinds.length); |
---|
| 226 | if (pickWorst) |
---|
| 227 | { |
---|
| 228 | if( bucketValues[j]>bucketValues[i] ) { i = j; si = 0; } |
---|
| 229 | else if( bucketValues[i]>bucketValues[j] ) { } // do nothing |
---|
| 230 | else |
---|
| 231 | { |
---|
| 232 | if (si==0) |
---|
| 233 | si = oldinds[i].size(); |
---|
| 234 | long sj = oldinds[j].size(); |
---|
| 235 | |
---|
| 236 | if (sj >= si) // sj's got worse lookin' trees |
---|
| 237 | { i = j; si = sj; } |
---|
| 238 | } |
---|
| 239 | } |
---|
| 240 | else |
---|
| 241 | { |
---|
| 242 | if( bucketValues[j]<bucketValues[i] ) { i = j; si = 0; } |
---|
| 243 | else if( bucketValues[i]<bucketValues[j] ) { } // do nothing |
---|
| 244 | else |
---|
| 245 | { |
---|
| 246 | if (si==0) |
---|
| 247 | si = oldinds[i].size(); |
---|
| 248 | long sj = oldinds[j].size(); |
---|
| 249 | |
---|
| 250 | if (sj < si) // sj's got better lookin' trees |
---|
| 251 | { i = j; si = sj; } |
---|
| 252 | } |
---|
| 253 | } |
---|
| 254 | } |
---|
| 255 | return i; |
---|
| 256 | } |
---|
| 257 | |
---|
| 258 | public void individualReplaced(final SteadyStateEvolutionState state, |
---|
| 259 | final int subpopulation, |
---|
| 260 | final int thread, |
---|
| 261 | final int individual) |
---|
| 262 | { return; } |
---|
| 263 | |
---|
| 264 | public void sourcesAreProperForm(final SteadyStateEvolutionState state) |
---|
| 265 | { return; } |
---|
| 266 | |
---|
| 267 | } |
---|