1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
|
package de.lmu.ifi.dbs.elki.algorithm.clustering.kmeans.initialization;
/*
This file is part of ELKI:
Environment for Developing KDD-Applications Supported by Index-Structures
Copyright (C) 2015
Ludwig-Maximilians-Universität München
Lehr- und Forschungseinheit für Datenbanksysteme
ELKI Development Team
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Affero General Public License for more details.
You should have received a copy of the GNU Affero General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
import java.util.ArrayList;
import java.util.List;
import de.lmu.ifi.dbs.elki.data.NumberVector;
import de.lmu.ifi.dbs.elki.database.Database;
import de.lmu.ifi.dbs.elki.database.datastore.DataStoreFactory;
import de.lmu.ifi.dbs.elki.database.datastore.DataStoreUtil;
import de.lmu.ifi.dbs.elki.database.datastore.WritableDoubleDataStore;
import de.lmu.ifi.dbs.elki.database.ids.ArrayModifiableDBIDs;
import de.lmu.ifi.dbs.elki.database.ids.DBIDIter;
import de.lmu.ifi.dbs.elki.database.ids.DBIDRef;
import de.lmu.ifi.dbs.elki.database.ids.DBIDUtil;
import de.lmu.ifi.dbs.elki.database.ids.DBIDVar;
import de.lmu.ifi.dbs.elki.database.ids.DBIDs;
import de.lmu.ifi.dbs.elki.database.query.distance.DistanceQuery;
import de.lmu.ifi.dbs.elki.database.relation.Relation;
import de.lmu.ifi.dbs.elki.distance.distancefunction.NumberVectorDistanceFunction;
import de.lmu.ifi.dbs.elki.math.random.RandomFactory;
/**
* K-Means initialization by repeatedly choosing the farthest point (by the
* <em>sum</em> of distances to previous objects).
*
* Note: this is less random than other initializations, so running multiple
* times will be more likely to return the same local minima.
*
* @author Erich Schubert
*
* @param <O> Object type for kmedoids and kmedians
*/
public class FarthestSumPointsInitialMeans<O> extends FarthestPointsInitialMeans<O> {
/**
* Constructor.
*
* @param rnd Random generator.
* @param dropfirst Flag to discard the first vector.
*/
public FarthestSumPointsInitialMeans(RandomFactory rnd, boolean dropfirst) {
super(rnd, dropfirst);
}
@Override
public <T extends NumberVector, V extends NumberVector> List<V> chooseInitialMeans(Database database, Relation<T> relation, int k, NumberVectorDistanceFunction<? super T> distanceFunction, NumberVector.Factory<V> factory) {
// Get a distance query
DistanceQuery<T> distQ = database.getDistanceQuery(relation, distanceFunction);
DBIDs ids = relation.getDBIDs();
WritableDoubleDataStore store = DataStoreUtil.makeDoubleStorage(ids, DataStoreFactory.HINT_HOT | DataStoreFactory.HINT_TEMP, 0.);
// Chose first mean
List<V> means = new ArrayList<>(k);
DBIDRef first = DBIDUtil.randomSample(ids, rnd);
T prevmean = relation.get(first);
means.add(factory.newNumberVector(prevmean));
// Find farthest object each.
DBIDVar best = DBIDUtil.newVar(first);
for(int i = (dropfirst ? 0 : 1); i < k; i++) {
double maxdist = Double.NEGATIVE_INFINITY;
for(DBIDIter it = ids.iter(); it.valid(); it.advance()) {
final double prev = store.doubleValue(it);
if(prev != prev) {
continue; // NaN: already chosen!
}
double dsum = prev + distQ.distance(prevmean, it);
// Don't store distance to first mean, when it will be dropped below.
if(i > 0) {
store.putDouble(it, dsum);
}
if(dsum > maxdist) {
maxdist = dsum;
best.set(it);
}
}
// Add new mean (and drop the initial mean when desired)
if(i == 0) {
means.clear(); // Remove temporary first element.
}
store.putDouble(best, Double.NaN); // So it won't be chosen twice.
prevmean = relation.get(best);
means.add(factory.newNumberVector(prevmean));
}
// Explicitly destroy temporary data.
store.destroy();
return means;
}
@Override
public DBIDs chooseInitialMedoids(int k, DBIDs ids, DistanceQuery<? super O> distQ) {
@SuppressWarnings("unchecked")
final Relation<O> relation = (Relation<O>) distQ.getRelation();
WritableDoubleDataStore store = DataStoreUtil.makeDoubleStorage(ids, DataStoreFactory.HINT_HOT | DataStoreFactory.HINT_TEMP, 0.);
ArrayModifiableDBIDs means = DBIDUtil.newArray(k);
DBIDRef first = DBIDUtil.randomSample(ids, rnd);
means.add(first);
DBIDVar prevmean = DBIDUtil.newVar(first);
DBIDVar best = DBIDUtil.newVar(first);
for(int i = (dropfirst ? 0 : 1); i < k; i++) {
// Find farthest object:
double maxdist = Double.NEGATIVE_INFINITY;
for(DBIDIter it = relation.iterDBIDs(); it.valid(); it.advance()) {
final double prev = store.doubleValue(it);
if(prev != prev) {
continue; // NaN: already chosen!
}
double dsum = prev + distQ.distance(prevmean, it);
// Don't store distance to first mean, when it will be dropped below.
if(i > 0) {
store.putDouble(it, dsum);
}
if(dsum > maxdist) {
maxdist = dsum;
best.set(it);
}
}
// Add new mean:
if(i == 0) {
means.clear(); // Remove temporary first element.
}
store.putDouble(best, Double.NaN); // So it won't be chosen twice.
prevmean.set(best);
means.add(best);
}
store.destroy();
return means;
}
/**
* Parameterization class.
*
* @author Erich Schubert
*
* @apiviz.exclude
*/
public static class Parameterizer<V> extends FarthestPointsInitialMeans.Parameterizer<V> {
/**
* Flag for discarding the first object chosen.
*/
protected boolean keepfirst = false;
@Override
protected FarthestSumPointsInitialMeans<V> makeInstance() {
return new FarthestSumPointsInitialMeans<>(rnd, !keepfirst);
}
}
}
|
