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
|
package de.lmu.ifi.dbs.elki.datasource.filter.normalization.instancewise;
/*
This file is part of ELKI:
Environment for Developing KDD-Applications Supported by Index-Structures
Copyright (C) 2014
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 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.VectorTypeInformation;
import de.lmu.ifi.dbs.elki.datasource.filter.normalization.AbstractStreamNormalization;
import de.lmu.ifi.dbs.elki.utilities.optionhandling.AbstractParameterizer;
/**
* Normalize vectors such that they have zero mean and unit variance.
*
* @author Erich Schubert
*
* @param <V> vector type
*/
public class InstanceMeanVarianceNormalization<V extends NumberVector> extends AbstractStreamNormalization<V> {
/**
* Multiplicity of the vector.
*/
private int multiplicity;
/**
* Constructor.
*/
public InstanceMeanVarianceNormalization() {
super();
}
@Override
protected V filterSingleObject(V featureVector) {
double[] raw = featureVector.getColumnVector().getArrayRef();
if(raw.length == 0) {
return factory.newNumberVector(new double[] {});
}
if(raw.length == 1) {
// Constant, but preserve NaNs
return factory.newNumberVector(new double[] { raw[0] == raw[0] ? 0. : Double.NaN });
}
// Multivariate codepath:
if(multiplicity > 1) {
assert (raw.length % multiplicity == 0) : "Vector length is not divisible by multiplicity?";
return factory.newNumberVector(multivariateStandardization(raw));
}
return factory.newNumberVector(univariateStandardization(raw));
}
protected double[] univariateStandardization(double[] raw) {
// Two pass normalization is numerically most stable,
// And Java should optimize this well enough.
double sum = 0.;
for(int i = 0; i < raw.length; ++i) {
final double v = raw[i];
if(v != v) { // NaN guard
continue;
}
sum += v;
}
final double mean = sum / raw.length;
double ssum = 0.;
for(int i = 0; i < raw.length; ++i) {
double v = raw[i] - mean;
if(v != v) {
continue;
}
ssum += v * v;
}
final double std = Math.sqrt(ssum) / (raw.length - 1);
if(std > 0.) {
for(int i = 0; i < raw.length; ++i) {
raw[i] = (raw[i] - mean) / std;
}
}
return raw;
}
protected double[] multivariateStandardization(double[] raw) {
final int len = raw.length / multiplicity;
if(len <= 1) {
return raw;
}
// Two pass normalization is numerically most stable,
// And Java should optimize this well enough.
double[] mean = new double[multiplicity];
for(int i = 0, j = 0; i < raw.length; ++i, j = ++j % multiplicity) {
final double v = raw[i];
if(v != v) { // NaN guard
continue;
}
mean[j] += v;
}
for(int j = 0; j < multiplicity; ++j) {
mean[j] /= len;
}
double[] std = new double[multiplicity];
for(int i = 0, j = 0; i < raw.length; ++i, j = ++j % multiplicity) {
double v = raw[i] - mean[j];
if(v != v) {
continue;
}
std[j] += v * v;
}
for(int j = 0; j < multiplicity; ++j) {
std[j] = std[j] > 0. ? Math.sqrt(std[j]) / (len - 1) : 1;
}
for(int i = 0, j = 0; i < raw.length; ++i, j = ++j % multiplicity) {
raw[i] = (raw[i] - mean[j]) / std[j];
}
return raw;
}
@Override
protected void initializeOutputType(SimpleTypeInformation<V> type) {
super.initializeOutputType(type);
multiplicity = ((VectorTypeInformation<?>) type).getMultiplicity();
}
@Override
protected SimpleTypeInformation<? super V> getInputTypeRestriction() {
return TypeUtil.NUMBER_VECTOR_VARIABLE_LENGTH;
}
/**
* Parameterization class.
*
* @author Erich Schubert
*
* @apiviz.exclude
*/
public static class Parameterizer<V extends NumberVector> extends AbstractParameterizer {
@Override
protected InstanceMeanVarianceNormalization<V> makeInstance() {
return new InstanceMeanVarianceNormalization<>();
}
}
}
|
