de.lmu.ifi.dbs.elki.datasource.filter.normalization.columnwise

Class AttributeWiseMADNormalization<V extends NumberVector>

java.lang.Object

de.lmu.ifi.dbs.elki.datasource.filter.normalization.columnwise.AttributeWiseMADNormalization<V>

Type Parameters:

V - vector type

All Implemented Interfaces:

Normalization<V>, ObjectFilter

@Alias(value="de.lmu.ifi.dbs.elki.datasource.filter.normalization.AttributeWiseMADNormalization")
public class AttributeWiseMADNormalization<V extends NumberVector>
extends Object
implements Normalization<V>

Median Absolute Deviation is used for scaling the data set as follows: First, the median, and median absolute deviation are computed in each axis. Then, each value is projected to (x - median(X)) / MAD(X). This is similar to z-standardization of data sets, except that it is more robust towards outliers, and only slightly more expensive to compute.

Since:

0.6.0

Author:

Erich Schubert

Field Summary

Fields

Modifier and Type	Field and Description
protected NumberVector.Factory<V>	factory Number vector factory.
private double[]	imadsigma Stores the inverse median absolute deviation in each dimension.
private static Logging	LOG Class logger.
private double[]	median Stores the median in each dimension.

Constructor Summary

Constructors

Constructor and Description
AttributeWiseMADNormalization() Constructor.

Method Summary

Methods

Modifier and Type	Method and Description
MultipleObjectsBundle	filter(MultipleObjectsBundle objects) Filter a set of object packages.
private double	normalize(int d, double val) Normalize a single dimension.
private double	restore(int d, double val) Restore a single dimension.
V	restore(V featureVector) Transforms a feature vector to the original attribute ranges.
String	toString()
LinearEquationSystem	transform(LinearEquationSystem linearEquationSystem) Transforms a linear equation system describing linear dependencies derived on the normalized space into a linear equation system describing linear dependencies quantitatively adapted to the original space.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Field Detail

LOG

private static final Logging LOG

Class logger.

factory

protected NumberVector.Factory<V extends NumberVector> factory

Number vector factory.

median

private double[] median

Stores the median in each dimension.

imadsigma

private double[] imadsigma

Stores the inverse median absolute deviation in each dimension.

Constructor Detail

AttributeWiseMADNormalization

public AttributeWiseMADNormalization()

Constructor.

Method Detail

filter

public MultipleObjectsBundle filter(MultipleObjectsBundle objects)

Description copied from interface: ObjectFilter

Filter a set of object packages.

Specified by:

filter in interface ObjectFilter

Parameters:

objects - Object to filter

Returns:

Filtered objects

restore

public V restore(V featureVector)
                               throws NonNumericFeaturesException

Description copied from interface: Normalization

Transforms a feature vector to the original attribute ranges.

Specified by:

restore in interface Normalization<V extends NumberVector>

Parameters:

featureVector - a feature vector to be transformed into original space

Returns:

a feature vector transformed into original space corresponding to the given feature vector

Throws:

NonNumericFeaturesException - feature vector is not compatible with values initialized during normalization

transform

public LinearEquationSystem transform(LinearEquationSystem linearEquationSystem)
                               throws NonNumericFeaturesException

Description copied from interface: Normalization

Transforms a linear equation system describing linear dependencies derived on the normalized space into a linear equation system describing linear dependencies quantitatively adapted to the original space.

Specified by:

transform in interface Normalization<V extends NumberVector>

Parameters:

linearEquationSystem - the linear equation system to be transformed

Returns:

a linear equation system describing linear dependencies derived on the normalized space transformed into a linear equation system describing linear dependencies quantitatively adapted to the original space

Throws:

NonNumericFeaturesException - if specified linear equation system is not compatible with values initialized during normalization

normalize

private double normalize(int d,
               double val)

Normalize a single dimension.

Parameters:

d - Dimension

val - Value

Returns:

Normalized value

restore

private double restore(int d,
             double val)

Restore a single dimension.

Parameters:

d - Dimension

val - Value

Returns:

Normalized value

toString

public String toString()

Overrides:

toString in class Object