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
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
|
package de.lmu.ifi.dbs.elki.datasource.filter.transform;
/*
This file is part of ELKI:
Environment for Developing KDD-Applications Supported by Index-Structures
Copyright (C) 2013
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.List;
import de.lmu.ifi.dbs.elki.data.NumberVector;
import de.lmu.ifi.dbs.elki.data.type.SimpleTypeInformation;
import de.lmu.ifi.dbs.elki.data.type.TypeUtil;
import de.lmu.ifi.dbs.elki.data.type.VectorFieldTypeInformation;
import de.lmu.ifi.dbs.elki.datasource.filter.AbstractVectorConversionFilter;
import de.lmu.ifi.dbs.elki.logging.Logging;
import de.lmu.ifi.dbs.elki.math.linearalgebra.CovarianceMatrix;
import de.lmu.ifi.dbs.elki.math.linearalgebra.EigenPair;
import de.lmu.ifi.dbs.elki.math.linearalgebra.SortedEigenPairs;
import de.lmu.ifi.dbs.elki.math.linearalgebra.VMath;
import de.lmu.ifi.dbs.elki.math.linearalgebra.pca.EigenPairFilter;
import de.lmu.ifi.dbs.elki.math.linearalgebra.pca.PCAResult;
import de.lmu.ifi.dbs.elki.math.linearalgebra.pca.PCARunner;
import de.lmu.ifi.dbs.elki.utilities.Alias;
import de.lmu.ifi.dbs.elki.utilities.exceptions.AbortException;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.AbstractParameterizer;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.OptionID;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.parameterization.Parameterization;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.parameters.ObjectParameter;
/**
* Apply principal component analysis to the data set.
*
* This process is also known as "Whitening transformation".
*
* If you want to also reduce dimensionality, set the
* {@link Parameterizer#FILTER_ID} parameter!
*
* @author Erich Schubert
*
* @param <O> Vector type
*/
@Alias({ "whiten", "whitening", "pca" })
public class GlobalPrincipalComponentAnalysisTransform<O extends NumberVector<?>> extends AbstractVectorConversionFilter<O, O> {
/**
* Class logger.
*/
private static final Logging LOG = Logging.getLogger(GlobalPrincipalComponentAnalysisTransform.class);
/**
* Filter to use for dimensionality reduction.
*/
EigenPairFilter filter = null;
/**
* Actual dataset dimensionality.
*/
int dim = -1;
/**
* Covariance matrix builder.
*/
CovarianceMatrix covmat = null;
/**
* Final projection after analysis run.
*/
double[][] proj = null;
/**
* Projection buffer.
*/
double[] buf = null;
/**
* Vector for data set centering.
*/
double[] mean = null;
/**
* Constructor.
*
* @param filter Filter to use for dimensionality reduction.
*/
public GlobalPrincipalComponentAnalysisTransform(EigenPairFilter filter) {
super();
this.filter = filter;
}
@Override
protected boolean prepareStart(SimpleTypeInformation<O> in) {
if (!(in instanceof VectorFieldTypeInformation)) {
throw new AbortException("PCA can only applied to fixed dimensionality vectors");
}
dim = ((VectorFieldTypeInformation<?>) in).getDimensionality();
covmat = new CovarianceMatrix(dim);
return true;
}
@Override
protected void prepareProcessInstance(O obj) {
covmat.put(obj);
}
@Override
protected void prepareComplete() {
mean = covmat.getMeanVector().getArrayRef();
PCAResult pcares = (new PCARunner<O>(null)).processCovarMatrix(covmat.destroyToSampleMatrix());
SortedEigenPairs eps = pcares.getEigenPairs();
covmat = null;
if (filter == null) {
proj = new double[dim][dim];
for (int d = 0; d < dim; d++) {
EigenPair ep = eps.getEigenPair(d);
double[] ev = ep.getEigenvector().getArrayRef();
double eval = Math.sqrt(ep.getEigenvalue());
// Fill weighted and transposed:
for (int i = 0; i < dim; i++) {
proj[d][i] = ev[i] / eval;
}
}
} else {
List<EigenPair> axes = filter.filter(eps).getStrongEigenPairs();
final int pdim = axes.size(); // Projection dimensionality
if (LOG.isVerbose()) {
LOG.verbose("Reducing dimensionality from " + dim + " to " + pdim + " via PCA.");
}
proj = new double[pdim][dim];
for (int d = 0; d < pdim; d++) {
EigenPair ep = axes.get(d);
double[] ev = ep.getEigenvector().getArrayRef();
double eval = Math.sqrt(ep.getEigenvalue());
// Fill weighted and transposed:
for (int i = 0; i < dim; i++) {
proj[d][i] = ev[i] / eval;
}
}
}
buf = new double[dim];
}
@Override
protected O filterSingleObject(O obj) {
// Shift by mean and copy
for (int i = 0; i < dim; i++) {
buf[i] = obj.doubleValue(i) - mean[i];
}
double[] p = VMath.times(proj, buf);
return factory.newNumberVector(p);
}
@Override
protected SimpleTypeInformation<? super O> getInputTypeRestriction() {
return TypeUtil.NUMBER_VECTOR_FIELD;
}
@Override
protected SimpleTypeInformation<? super O> convertedType(SimpleTypeInformation<O> in) {
initializeOutputType(in);
return new VectorFieldTypeInformation<>(factory, proj.length);
}
@Override
protected Logging getLogger() {
return LOG;
}
/**
* Parameterization class.
*
* @author Erich Schubert
*
* @apiviz.exclude
*/
public static class Parameterizer<O extends NumberVector<?>> extends AbstractParameterizer {
/**
* To specify the eigenvectors to keep.
*/
public static final OptionID FILTER_ID = new OptionID("globalpca.filter", "Filter to use for dimensionality reduction.");
/**
* Filter to use for dimensionality reduction.
*/
EigenPairFilter filter = null;
@Override
protected void makeOptions(Parameterization config) {
super.makeOptions(config);
ObjectParameter<EigenPairFilter> filterP = new ObjectParameter<>(FILTER_ID, EigenPairFilter.class, true);
if (config.grab(filterP)) {
filter = filterP.instantiateClass(config);
}
}
@Override
protected GlobalPrincipalComponentAnalysisTransform<O> makeInstance() {
return new GlobalPrincipalComponentAnalysisTransform<>(filter);
}
}
}
|
