Abstract base class for machine learning classifiers. More...

#include <mitkAbstractClassifier.h>

Inheritance diagram for mitk::AbstractClassifier:

[legend]

Collaboration diagram for mitk::AbstractClassifier:

[legend]

Public Member Functions
	mitkClassMacro (AbstractClassifier, BaseData)

virtual void	Train (const Eigen::MatrixXd &X, const Eigen::MatrixXi &Y)=0
	Train the classifier on a labeled dataset. More...

virtual Eigen::MatrixXi	Predict (const Eigen::MatrixXd &X)=0
	Predict class labels for the given input samples. More...

Eigen::MatrixXi &	GetLabels ()
	Get the predicted label matrix from the last prediction. More...

virtual bool	SupportsPointWiseWeight ()=0
	Query whether this classifier supports per-sample weighting. More...

virtual Eigen::MatrixXd &	GetPointWiseWeight ()
	Get the per-sample weight matrix. More...

virtual void	SetPointWiseWeight (const Eigen::MatrixXd &W)
	Set the per-sample weight matrix. More...

virtual void	UsePointWiseWeight (bool value)
	Enable or disable per-sample weighting during training. More...

virtual bool	IsUsingPointWiseWeight ()
	Query whether per-sample weighting is currently enabled. More...

virtual bool	SupportsPointWiseProbability ()=0
	Query whether this classifier supports per-sample class probability output. More...

virtual Eigen::MatrixXd &	GetPointWiseProbabilities ()
	Get the per-sample class probability matrix from the last prediction. More...

virtual void	UsePointWiseProbability (bool value)
	Enable or disable per-sample class probability computation. More...

virtual bool	IsUsingPointWiseProbability ()
	Query whether per-sample class probability output is currently enabled. More...

void	SetNthItems (const char *val, unsigned int idx)
	Set an item in the property-based item list at the given index. More...

std::string	GetNthItems (unsigned int idx) const
	Get the item string at the given index from the property-based item list. More...

void	SetItemList (std::vector< std::string > list)
	Replace the entire item list with the given vector of strings. More...

std::vector< std::string >	GetItemList () const
	Retrieve all items from the property-based item list. More...

void	SetRequestedRegionToLargestPossibleRegion () override
	Set the RequestedRegion to the LargestPossibleRegion. More...

bool	RequestedRegionIsOutsideOfTheBufferedRegion () override
	Determine whether the RequestedRegion is outside of the BufferedRegion. More...

bool	VerifyRequestedRegion () override
	Verify that the RequestedRegion is within the LargestPossibleRegion. More...

void	SetRequestedRegion (const itk::DataObject *) override
	Set the requested region from this data object to match the requested region of the data object passed in as a parameter. More...

bool	IsEmpty () const override
	Check whether the object contains data at any time step. More...

Public Member Functions inherited from mitk::BaseData
	mitkClassMacroItkParent (BaseData, itk::DataObject)

BaseProperty::ConstPointer	GetConstProperty (const std::string &propertyKey, const std::string &contextName="", bool fallBackOnDefaultContext=true) const override
	Get a const property by its key. More...

std::vector< std::string >	GetPropertyKeys (const std::string &contextName="", bool includeDefaultContext=false) const override
	Get all property keys stored in the property list. More...

std::vector< std::string >	GetPropertyContextNames () const override
	Get the names of all property contexts. More...

BaseProperty *	GetNonConstProperty (const std::string &propertyKey, const std::string &contextName="", bool fallBackOnDefaultContext=true) override
	Get a non-const property by its key. More...

void	SetProperty (const std::string &propertyKey, BaseProperty *property, const std::string &contextName="", bool fallBackOnDefaultContext=false) override
	Set a property in the property list. More...

void	RemoveProperty (const std::string &propertyKey, const std::string &contextName="", bool fallBackOnDefaultContext=false) override
	Remove a property from the property list. More...

const mitk::TimeGeometry *	GetTimeGeometry () const
	Return the TimeGeometry of the data as const pointer. More...

mitk::TimeGeometry *	GetTimeGeometry ()
	Return the TimeGeometry of the data as non-const pointer. More...

const mitk::TimeGeometry *	GetUpdatedTimeGeometry ()
	Return the TimeGeometry of the data after ensuring it is up-to-date. More...

virtual void	Expand (unsigned int timeSteps)
	Expand the TimeGeometry to a number of time steps. More...

const mitk::BaseGeometry *	GetUpdatedGeometry (int t=0)
	Return the BaseGeometry of the data at time step t after ensuring it is up-to-date. More...

mitk::BaseGeometry *	GetGeometry (int t=0) const
	Return the BaseGeometry of the data at time step t as a non-const pointer. More...

void	UpdateOutputInformation () override
	Update the information for this BaseData so that it can be used as an output of a BaseProcess. More...

void	CopyInformation (const itk::DataObject *data) override
	Copy information from the specified data set. More...

virtual bool	IsInitialized () const
	Check whether the data has been initialized. More...

virtual void	Clear ()
	Reset the data object by calling ClearData() and InitializeEmpty(). More...

virtual bool	IsEmptyTimeStep (unsigned int t) const
	Check whether the object contains data at the specified time step. More...

void	ExecuteOperation (Operation *operation) override
	Execute an operation on this data object. More...

virtual void	SetGeometry (BaseGeometry *aGeometry3D)
	Set the BaseGeometry of the data, which will be referenced (not copied!). More...

virtual void	SetTimeGeometry (TimeGeometry *geometry)
	Set the TimeGeometry of the data, which will be referenced (not copied!). More...

virtual void	SetClonedGeometry (const BaseGeometry *aGeometry3D)
	Set a clone of the provided geometry as the BaseGeometry of the data. More...

virtual void	SetClonedTimeGeometry (const TimeGeometry *geometry)
	Set a clone of the provided TimeGeometry as the TimeGeometry of the data. More...

virtual void	SetClonedGeometry (const BaseGeometry *aGeometry3D, unsigned int time)
	Set a clone of the provided geometry as the BaseGeometry of a given time step. More...

mitk::PropertyList::Pointer	GetPropertyList () const
	Get the data's property list. More...

void	SetPropertyList (PropertyList *propertyList)
	Set the data's property list, replacing the existing one. More...

mitk::BaseProperty::Pointer	GetProperty (const char *propertyKey) const
	Get a property by its key from the PropertyList. More...

void	SetProperty (const char propertyKey, BaseProperty property)
	Set a property in the PropertyList. More...

virtual void	SetOrigin (const Point3D &origin)
	Convenience method for setting the origin of the BaseGeometry instances of all time steps. More...

itk::SmartPointer< mitk::BaseDataSource >	GetSource () const
	Get the process object that generated this data object. More...

unsigned int	GetTimeSteps () const
	Get the number of time steps from the TimeGeometry. More...

itk::ModifiedTimeType	GetMTime () const override
	Get the modification time of this data object or its geometry, whichever was modified most recently. More...

void	Graft (const DataObject *) override
	Graft data and information from another data object. More...

Public Member Functions inherited from mitk::OperationActor
	itkTypeMacroNoParent (OperationActor) virtual ~OperationActor()

Public Member Functions inherited from mitk::Identifiable
	Identifiable ()
	Default constructor. Generates a new random UID. More...

	Identifiable (const UIDType &uid)
	Construct with a pre-existing UID. More...

	Identifiable (const Identifiable &)=delete
	Copy construction is deleted to prevent UID duplication. More...

	Identifiable (Identifiable &&) noexcept
	Move constructor. Transfers ownership of the UID. More...

virtual	~Identifiable ()
	Virtual destructor. More...

Identifiable &	operator= (const Identifiable &)=delete
	Copy assignment is deleted to prevent UID duplication. More...

Identifiable &	operator= (Identifiable &&other) noexcept
	Move assignment operator. Transfers ownership of the UID. More...

virtual UIDType	GetUID () const
	Get the unique identifier of this object. More...

UIDType	GetRuntimeUID () const noexcept
	Get a runtime-unique ID for this object instance. More...

Public Member Functions inherited from mitk::IPropertyOwner
	~IPropertyOwner () override

virtual bool	PropertyIsOwned (const std::string &propertyKey, const std::string &contextName="", bool fallBackOnDefaultContext=true) const
	Checks if a certain property exists. More...

Public Member Functions inherited from mitk::IPropertyProvider
virtual	~IPropertyProvider ()
	Virtual destructor. More...

Protected Attributes
Eigen::MatrixXi	m_OutLabel

Eigen::MatrixXd	m_PointWiseWeight

bool	m_IsUsingPointWiseWeight

Eigen::MatrixXd	m_OutProbability

bool	m_IsUsingPointWiseProbability

Protected Attributes inherited from mitk::BaseData
bool	m_LastRequestedRegionWasOutsideOfTheBufferedRegion

unsigned int	m_SourceOutputIndexDuplicate

bool	m_Initialized

Additional Inherited Members
Public Types inherited from mitk::Identifiable
using	UIDType = std::string
	Type alias for unique identifiers. More...

Protected Member Functions inherited from mitk::BaseData
	BaseData ()

	BaseData (const BaseData &other)

	~BaseData () override

virtual void	InitializeTimeGeometry (unsigned int timeSteps=1)
	Initialize the TimeGeometry for a number of time steps. The TimeGeometry is initialized empty and evenly timed. In many cases it will be necessary to overwrite this in sub-classes. More...

virtual void	ClearData ()
	reset to non-initialized state, release memory More...

virtual void	InitializeEmpty ()
	Pure virtual; Must be used in subclasses to get a data object to a valid state. Should at least create one empty object and call Superclass::InitializeTimeGeometry() to ensure an existing valid geometry. More...

void	PrintSelf (std::ostream &os, itk::Indent indent) const override

Protected Member Functions inherited from mitk::Identifiable
virtual void	SetUID (const UIDType &uid)

Detailed Description

Abstract base class for machine learning classifiers.

This class defines a common interface for training and prediction with classifiers that operate on Eigen matrices. Derived classes must implement training, prediction, and declare whether they support sample-level weighting and probability estimation.

Features include:

Training from feature matrix X and label vector Y
Prediction returning class labels
Optional per-sample weighting
Optional per-sample class probability output
An item list stored via the property system for configuration

See also: AbstractGlobalImageFeature

Definition at line 49 of file mitkAbstractClassifier.h.

Member Function Documentation

◆ GetItemList()

std::vector<std::string> mitk::AbstractClassifier::GetItemList ( ) const

Retrieve all items from the property-based item list.

Returns: A vector of strings in index order until a gap is found.

◆ GetLabels()

Eigen::MatrixXi& mitk::AbstractClassifier::GetLabels ( )

inline

Get the predicted label matrix from the last prediction.

Returns: Reference to the label matrix of shape [n_samples, 1].

Definition at line 76 of file mitkAbstractClassifier.h.

◆ GetNthItems()

std::string mitk::AbstractClassifier::GetNthItems ( unsigned int idx ) const

Get the item string at the given index from the property-based item list.

Parameters

[in] idx Zero-based index position.

Returns: The string value at the given index.

◆ GetPointWiseProbabilities()

virtual Eigen::MatrixXd& mitk::AbstractClassifier::GetPointWiseProbabilities ( )

inlinevirtual

Get the per-sample class probability matrix from the last prediction.

Returns: Reference to the probability matrix.

Definition at line 158 of file mitkAbstractClassifier.h.

◆ GetPointWiseWeight()

virtual Eigen::MatrixXd& mitk::AbstractClassifier::GetPointWiseWeight ( )

inlinevirtual

Get the per-sample weight matrix.

Returns: Reference to the weight matrix of shape [n_samples, 1].

Definition at line 102 of file mitkAbstractClassifier.h.

◆ IsEmpty()

bool mitk::AbstractClassifier::IsEmpty ( ) const

inlineoverridevirtual

Check whether the object contains data at any time step.

Iterates over all time steps and returns true only if every time step is empty (or the object is not initialized).

Warning: Returns IsInitialized()==false by default for compatibility reasons. Override in sub-classes that support distinction between empty/non-empty state.

Returns: True if the data object is empty at all time steps.

See also: IsEmptyTimeStep

Reimplemented from mitk::BaseData.

Definition at line 234 of file mitkAbstractClassifier.h.

◆ IsUsingPointWiseProbability()

virtual bool mitk::AbstractClassifier::IsUsingPointWiseProbability ( )

inlinevirtual

Query whether per-sample class probability output is currently enabled.

Returns: True if pointwise probability output is active.

Definition at line 178 of file mitkAbstractClassifier.h.

◆ IsUsingPointWiseWeight()

virtual bool mitk::AbstractClassifier::IsUsingPointWiseWeight ( )

inlinevirtual

Query whether per-sample weighting is currently enabled.

Returns: True if pointwise weighting is active.

Definition at line 132 of file mitkAbstractClassifier.h.

◆ mitkClassMacro()

mitk::AbstractClassifier::mitkClassMacro	(	AbstractClassifier	,
		BaseData
	)

◆ Predict()

virtual Eigen::MatrixXi mitk::AbstractClassifier::Predict ( const Eigen::MatrixXd & X )

pure virtual

Predict class labels for the given input samples.

Parameters

[in] X Input samples. Matrix of shape [n_samples, n_features].

Returns: Predicted class labels. Matrix of shape [n_samples, 1].

◆ RequestedRegionIsOutsideOfTheBufferedRegion()

bool mitk::AbstractClassifier::RequestedRegionIsOutsideOfTheBufferedRegion ( )

inlineoverridevirtual

Determine whether the RequestedRegion is outside of the BufferedRegion.

This method returns true if the RequestedRegion is outside the BufferedRegion (true if at least one pixel is outside). This is used by the pipeline mechanism to determine whether a filter needs to re-execute in order to satisfy the current request. If the current RequestedRegion is already inside the BufferedRegion from the previous execution (and the current filter is up to date), then a given filter does not need to re-execute.

Returns: True if the requested region is outside the buffered region.

Implements mitk::BaseData.

Definition at line 227 of file mitkAbstractClassifier.h.

◆ SetItemList()

void mitk::AbstractClassifier::SetItemList ( std::vector< std::string > list )

Replace the entire item list with the given vector of strings.

Parameters

[in] list A vector of string items to store.

◆ SetNthItems()

void mitk::AbstractClassifier::SetNthItems	(	const char *	val,
		unsigned int	idx
	)

Set an item in the property-based item list at the given index.

Items are stored as string properties under keys "itemlist.0", "itemlist.1", etc.

Parameters

[in]	val	The string value to store.
[in]	idx	Zero-based index position.

◆ SetPointWiseWeight()

virtual void mitk::AbstractClassifier::SetPointWiseWeight ( const Eigen::MatrixXd & W )

inlinevirtual

Set the per-sample weight matrix.

Parameters

[in] W Weight matrix of shape [n_samples, 1].

Definition at line 112 of file mitkAbstractClassifier.h.

◆ SetRequestedRegion()

void mitk::AbstractClassifier::SetRequestedRegion ( const itk::DataObject * data )

inlineoverridevirtual

Set the requested region from this data object to match the requested region of the data object passed in as a parameter.

This method is implemented in the concrete subclasses of BaseData.

Parameters

[in] data The data object whose requested region will be matched.

Implements mitk::BaseData.

Definition at line 230 of file mitkAbstractClassifier.h.

◆ SetRequestedRegionToLargestPossibleRegion()

void mitk::AbstractClassifier::SetRequestedRegionToLargestPossibleRegion ( )

inlineoverridevirtual

Set the RequestedRegion to the LargestPossibleRegion.

This forces a filter to produce all of the output in one execution (i.e., not streaming) on the next call to Update().

Implements mitk::BaseData.

Definition at line 226 of file mitkAbstractClassifier.h.

◆ SupportsPointWiseProbability()

virtual bool mitk::AbstractClassifier::SupportsPointWiseProbability ( )

pure virtual

Query whether this classifier supports per-sample class probability output.

Returns: True if pointwise probability estimation is supported, false otherwise.

◆ SupportsPointWiseWeight()

virtual bool mitk::AbstractClassifier::SupportsPointWiseWeight ( )

pure virtual

Query whether this classifier supports per-sample weighting.

Returns: True if pointwise weighting is supported, false otherwise.

◆ Train()

virtual void mitk::AbstractClassifier::Train	(	const Eigen::MatrixXd &	X,
		const Eigen::MatrixXi &	Y
	)

pure virtual

Train the classifier on a labeled dataset.

Parameters

[in]	X	Training input samples. Matrix of shape [n_samples, n_features].
[in]	Y	Target class labels. Matrix of shape [n_samples, 1].

◆ UsePointWiseProbability()

virtual void mitk::AbstractClassifier::UsePointWiseProbability ( bool value )

inlinevirtual

Enable or disable per-sample class probability computation.

Parameters

[in] value True to enable probability output, false to disable.

Definition at line 168 of file mitkAbstractClassifier.h.

◆ UsePointWiseWeight()

virtual void mitk::AbstractClassifier::UsePointWiseWeight ( bool value )

inlinevirtual

Enable or disable per-sample weighting during training.

Parameters

[in] value True to enable weighting, false to disable.

Definition at line 122 of file mitkAbstractClassifier.h.

◆ VerifyRequestedRegion()

bool mitk::AbstractClassifier::VerifyRequestedRegion ( )

inlineoverridevirtual

Verify that the RequestedRegion is within the LargestPossibleRegion.

If the RequestedRegion is not within the LargestPossibleRegion, then the filter cannot possibly satisfy the request. This method returns true if the request can be satisfied (even if it will be necessary to process the entire LargestPossibleRegion) and returns false otherwise. Used by PropagateRequestedRegion(), which throws an InvalidRequestedRegionError exception if the requested region is not within the LargestPossibleRegion.

Returns: True if the requested region is within the largest possible region.

Implements mitk::BaseData.

Definition at line 228 of file mitkAbstractClassifier.h.

Member Data Documentation

◆ m_IsUsingPointWiseProbability

bool mitk::AbstractClassifier::m_IsUsingPointWiseProbability

protected

Definition at line 185 of file mitkAbstractClassifier.h.

◆ m_IsUsingPointWiseWeight

bool mitk::AbstractClassifier::m_IsUsingPointWiseWeight

protected

Definition at line 139 of file mitkAbstractClassifier.h.

◆ m_OutLabel

Eigen::MatrixXi mitk::AbstractClassifier::m_OutLabel

protected

Definition at line 82 of file mitkAbstractClassifier.h.

◆ m_OutProbability

Eigen::MatrixXd mitk::AbstractClassifier::m_OutProbability

protected

Definition at line 184 of file mitkAbstractClassifier.h.

◆ m_PointWiseWeight

Eigen::MatrixXd mitk::AbstractClassifier::m_PointWiseWeight

protected

Definition at line 138 of file mitkAbstractClassifier.h.

The documentation for this class was generated from the following file:

mitkAbstractClassifier.h

Public Member Functions

Protected Attributes

Additional Inherited Members

Detailed Description

Member Function Documentation

◆ GetItemList()

◆ GetLabels()

◆ GetNthItems()

◆ GetPointWiseProbabilities()

◆ GetPointWiseWeight()

◆ IsEmpty()

◆ IsUsingPointWiseProbability()

◆ IsUsingPointWiseWeight()

◆ mitkClassMacro()

◆ Predict()

◆ RequestedRegionIsOutsideOfTheBufferedRegion()

◆ SetItemList()

◆ SetNthItems()

◆ SetPointWiseWeight()

◆ SetRequestedRegion()

◆ SetRequestedRegionToLargestPossibleRegion()

◆ SupportsPointWiseProbability()

◆ SupportsPointWiseWeight()

◆ Train()

◆ UsePointWiseProbability()

◆ UsePointWiseWeight()

◆ VerifyRequestedRegion()

Member Data Documentation

◆ m_IsUsingPointWiseProbability

◆ m_IsUsingPointWiseWeight

◆ m_OutLabel

◆ m_OutProbability

◆ m_PointWiseWeight