weka.filters.unsupervised.attribute

Class Nystroem

java.lang.Object

weka.filters.Filter

weka.filters.SimpleFilter

weka.filters.SimpleBatchFilter

weka.filters.unsupervised.attribute.Nystroem

All Implemented Interfaces:

java.io.Serializable, CapabilitiesHandler, CapabilitiesIgnorer, CommandlineRunnable, OptionHandler, RevisionHandler, TechnicalInformationHandler

public class Nystroem
extends SimpleBatchFilter
implements TechnicalInformationHandler

Implements the Nystroem method for feature extraction using a kernel function.

For more information on the algorithm, see

Tianbao Yang, Yu-Feng Li, Mehrdad Mahdavi, Rong Jin, Zhi-Hua Zhou: Nystr"{o}m Method vs Random Fourier Features: A Theoretical and Empirical Comparison. In: Proc 26th Annual Conference on Neural Information Processing Systems, 485--493, 2012.

BibTeX:

 @inproceedings{Yang2012,
    author = {Tianbao Yang and Yu-Feng Li and Mehrdad Mahdavi and Rong Jin and Zhi-Hua Zhou},
    booktitle = {Proc 26th Annual Conference on Neural Information Processing Systems},
    pages = {485--493},
    title = {Nystr"{o}m Method vs Random Fourier Features: A Theoretical and Empirical Comparison},
    year = {2012},
    URL = {http://papers.nips.cc/paper/4588-nystrom-method-vs-random-fourier-features-a-theoretical-and-empirical-comparison}
 }
 

Valid options are:

 -K <kernel specification>
  The kernel function to use.

 -F <filter specification>
  The filter to use, which should be a filter that takes a sample of instances.

 -use-svd
  Whether to use singular value decomposition instead of eigendecomposition.

 -output-debug-info
  If set, filter is run in debug mode and
  may output additional info to the console

 -do-not-check-capabilities
  If set, filter capabilities are not checked before filter is built
  (use with caution).

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected Filter	m_Filter The filter to use for sub sampling.
protected Kernel	m_Kernel The kernel function to use.
protected Instances	m_Sample Stores the sample used for the approximation.
protected boolean	m_useSVD Determines whether singular value decomposition is used instead of eigendecomposition.
protected no.uib.cipr.matrix.Matrix	m_WeightingMatrix Stores the weighting matrix.
static double	SMALL Constant to avoid division by zero.

Fields inherited from class weka.filters.Filter

m_Debug, m_DoNotCheckCapabilities, m_FirstBatchDone, m_InputRelAtts, m_InputStringAtts, m_NewBatch, m_OutputRelAtts, m_OutputStringAtts

Constructor Summary

Constructors

Constructor and Description
Nystroem() Default constructor.

Method Summary

Modifier and Type	Method and Description
boolean	allowAccessToFullInputFormat() Returns whether to allow the determineOutputFormat(Instances) method access to the full dataset rather than just the header.
protected Instances	determineOutputFormat(Instances inputFormat) Determines the output format for the data that is produced by this filter.
Capabilities	getCapabilities() Returns the Capabilities of this filter.
Filter	getFilter() Gets the filter that is used for sub sampling.
Kernel	getKernel() Returns the kernel to use
java.lang.String	getRevision() Returns the revision string.
TechnicalInformation	getTechnicalInformation() Returns a reference to the algorithm implemented by this class.
boolean	getUseSVD() Gets whether to use singular value decomposition instead of eigendecomposition.
java.lang.String	globalInfo() Provides information regarding this class.
static void	main(java.lang.String[] argv) Main method for testing this class.
protected Instances	process(Instances instances) Takes a batch of data and transforms it.
void	setFilter(Filter filter) Sets the filter to use for sub sampling.
void	setKernel(Kernel value) sets the kernel to use
void	setUseSVD(boolean flag) Sets whether to use singular value decomposition instead of eigendecomposition.

Methods inherited from class weka.filters.SimpleBatchFilter

batchFinished, hasImmediateOutputFormat, input

Methods inherited from class weka.filters.SimpleFilter

reset, setInputFormat

Methods inherited from class weka.filters.Filter

batchFilterFile, bufferInput, copyValues, copyValues, debugTipText, doNotCheckCapabilitiesTipText, filterFile, flushInput, getCapabilities, getCopyOfInputFormat, getDebug, getDoNotCheckCapabilities, getInputFormat, getOptions, getOutputFormat, initInputLocators, initOutputLocators, inputFormatPeek, isFirstBatchDone, isNewBatch, isOutputFormatDefined, listOptions, makeCopies, makeCopy, mayRemoveInstanceAfterFirstBatchDone, numPendingOutput, output, outputFormatPeek, outputPeek, postExecution, preExecution, push, push, resetQueue, run, runFilter, setDebug, setDoNotCheckCapabilities, setOptions, setOutputFormat, testInputFormat, toString, useFilter, wekaStaticWrapper

Methods inherited from class java.lang.Object

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

Field Detail

SMALL

public static double SMALL

Constant to avoid division by zero.

m_Filter

protected Filter m_Filter

The filter to use for sub sampling.

m_Kernel

protected Kernel m_Kernel

The kernel function to use.

m_useSVD

protected boolean m_useSVD

Determines whether singular value decomposition is used instead of eigendecomposition.

m_Sample

protected Instances m_Sample

Stores the sample used for the approximation.

m_WeightingMatrix

protected no.uib.cipr.matrix.Matrix m_WeightingMatrix

Stores the weighting matrix.

Constructor Detail

Nystroem

public Nystroem()

Default constructor. Sets filter to Resample filter without replacement and sample size percentage to 10%. The kernel is set to PolyKernel with default options.

Method Detail

getCapabilities

public Capabilities getCapabilities()

Returns the Capabilities of this filter.

Specified by:

getCapabilities in interface CapabilitiesHandler

Overrides:

getCapabilities in class Filter

Returns:

the capabilities of this object

See Also:

Capabilities

getUseSVD

@OptionMetadata(displayName="Use SVD and not eigendecomposition",
                description="Whether to use singular value decomposition instead of eigendecomposition.",
                displayOrder=3,
                commandLineParamName="use-svd",
                commandLineParamSynopsis="-use-svd",
                commandLineParamIsFlag=true)
public boolean getUseSVD()

Gets whether to use singular value decomposition instead of eigendecomposition.

Returns:

true if SVD is to be used

setUseSVD

public void setUseSVD(boolean flag)

Sets whether to use singular value decomposition instead of eigendecomposition.

Parameters:

flag - true if singular value decomposition is to be used.

getFilter

@OptionMetadata(displayName="Filter for sampling instances",
                description="The filter to use, which should be a filter that takes a sample of instances.",
                displayOrder=2,
                commandLineParamName="F",
                commandLineParamSynopsis="-F <filter specification>")
public Filter getFilter()

Gets the filter that is used for sub sampling.

Returns:

the filter

setFilter

public void setFilter(Filter filter)

Sets the filter to use for sub sampling.

Parameters:

filter - the filter to use

setKernel

public void setKernel(Kernel value)

sets the kernel to use

Parameters:

value - the kernel to use

getKernel

@OptionMetadata(displayName="Kernel function",
                description="The kernel function to use.",
                displayOrder=1,
                commandLineParamName="K",
                commandLineParamSynopsis="-K <kernel specification>")
public Kernel getKernel()

Returns the kernel to use

Returns:

the current kernel

globalInfo

public java.lang.String globalInfo()

Provides information regarding this class.

Specified by:

globalInfo in class SimpleFilter

Returns:

string describing the method that this class implements

getTechnicalInformation

public TechnicalInformation getTechnicalInformation()

Returns a reference to the algorithm implemented by this class.

Specified by:

getTechnicalInformation in interface TechnicalInformationHandler

Returns:

a reference to the algorithm implemented by this class

allowAccessToFullInputFormat

public boolean allowAccessToFullInputFormat()

Returns whether to allow the determineOutputFormat(Instances) method access to the full dataset rather than just the header.

Default implementation returns false.

Overrides:

allowAccessToFullInputFormat in class SimpleBatchFilter

Returns:

whether determineOutputFormat has access to the full input dataset

determineOutputFormat

protected Instances determineOutputFormat(Instances inputFormat)
                                   throws java.lang.Exception

Determines the output format for the data that is produced by this filter.

Specified by:

determineOutputFormat in class SimpleFilter

Parameters:

inputFormat - the input format to base the output format on

Returns:

the output format

Throws:

java.lang.Exception - if a problem occurs when the output format is generated

See Also:

SimpleFilter.hasImmediateOutputFormat(), Filter.batchFinished()

process

protected Instances process(Instances instances)
                     throws java.lang.Exception

Takes a batch of data and transforms it.

Specified by:

process in class SimpleFilter

Parameters:

instances - the data to process

Returns:

the processed instances

Throws:

java.lang.Exception - is thrown if a problem occurs

See Also:

Filter.batchFinished()

getRevision

public java.lang.String getRevision()

Returns the revision string.

Specified by:

getRevision in interface RevisionHandler

Overrides:

getRevision in class Filter

Returns:

the revision

main

public static void main(java.lang.String[] argv)

Main method for testing this class.

Parameters:

argv - should contain arguments to the filter: use -h for help