Simple interface to support vector machines for classification and regression (via LIBSVM). More...

#include <OpenMS/ML/SVM/SimpleSVM.h>

Inheritance diagram for SimpleSVM:

Collaboration diagram for SimpleSVM:

Classes
struct	Prediction
	SVM/SVR prediction result. More...

Public Types
typedef std::map< String, std::vector< double > >	PredictorMap
	Mapping from predictor name to vector of predictor values. More...

typedef std::map< String, std::pair< double, double > >	ScaleMap
	Mapping from predictor name to predictor min and max. More...

Public Member Functions
	SimpleSVM ()
	Default constructor. More...

	~SimpleSVM () override
	Destructor. More...

void	setup (PredictorMap &predictors, const std::map< Size, double > &outcomes, bool classification=true)
	Load data and train a model. More...

void	predict (std::vector< Prediction > &predictions, std::vector< Size > indexes=std::vector< Size >()) const
	Predict class labels or regression values (and probabilities). More...

void	predict (PredictorMap &predictors, std::vector< Prediction > &predictions) const
	Predict class labels or regression values (and probabilities). More...

void	getFeatureWeights (std::map< String, double > &feature_weights) const
	Get the weights used for features (predictors) in the SVM model. More...

void	writeXvalResults (const String &path) const
	Write cross-validation (parameter optimization) results to a CSV file. More...

const ScaleMap &	getScaling () const
	Get data range of predictors before scaling to [0, 1]. More...

Public Member Functions inherited from DefaultParamHandler
	DefaultParamHandler (const String &name)
	Constructor with name that is displayed in error messages. More...

	DefaultParamHandler (const DefaultParamHandler &rhs)
	Copy constructor. More...

virtual	~DefaultParamHandler ()
	Destructor. More...

DefaultParamHandler &	operator= (const DefaultParamHandler &rhs)
	Assignment operator. More...

virtual bool	operator== (const DefaultParamHandler &rhs) const
	Equality operator. More...

void	setParameters (const Param &param)
	Sets the parameters. More...

const Param &	getParameters () const
	Non-mutable access to the parameters. More...

const Param &	getDefaults () const
	Non-mutable access to the default parameters. More...

const String &	getName () const
	Non-mutable access to the name. More...

void	setName (const String &name)
	Mutable access to the name. More...

const std::vector< String > &	getSubsections () const
	Non-mutable access to the registered subsections. More...

Protected Types
typedef std::vector< std::vector< std::vector< double > > >	SVMPerformance
	Classification (or regression) performance for different param. combinations (C/gamma/p): More...

Protected Member Functions
void	clear_ ()

void	scaleData_ (PredictorMap &predictors)
	Scale predictor values to range 0-1. More...

void	convertData_ (const PredictorMap &predictors)
	Convert predictors to LIBSVM format. More...

std::tuple< double, double, double >	chooseBestParameters_ (bool higher_better) const
	Choose best SVM parameters based on cross-validation results. More...

void	optimizeParameters_ (bool classification)
	Run cross-validation to optimize SVM parameters. More...

Protected Member Functions inherited from DefaultParamHandler
virtual void	updateMembers_ ()
	This method is used to update extra member variables at the end of the setParameters() method. More...

void	defaultsToParam_ ()
	Updates the parameters after the defaults have been set in the constructor. More...

Static Protected Member Functions
static void	printNull_ (const char *)
	Dummy function to suppress LIBSVM output. More...

Protected Attributes
std::vector< std::vector< struct svm_node > >	nodes_
	Values of predictors (LIBSVM format) More...

struct svm_problem	data_
	SVM training data (LIBSVM format) More...

struct svm_parameter	svm_params_
	SVM parameters (LIBSVM format) More...

struct svm_model *	model_
	Pointer to SVM model (LIBSVM format) More...

std::vector< String >	predictor_names_
	Names of predictors in the model (excluding uninformative ones) More...

Size	n_parts_
	Number of partitions for cross-validation. More...

std::vector< double >	log2_C_
	Parameter values to try during optimization. More...

std::vector< double >	log2_gamma_

std::vector< double >	log2_p_

ScaleMap	scaling_
	Mapping from predictor name to predictor min and max. More...

SVMPerformance	performance_
	Cross-validation results. More...

Protected Attributes inherited from DefaultParamHandler
Param	param_
	Container for current parameters. More...

Param	defaults_
	Container for default parameters. This member should be filled in the constructor of derived classes! More...

std::vector< String >	subsections_
	Container for registered subsections. This member should be filled in the constructor of derived classes! More...

String	error_name_
	Name that is displayed in error messages during the parameter checking. More...

bool	check_defaults_
	If this member is set to false no checking if parameters in done;. More...

bool	warn_empty_defaults_
	If this member is set to false no warning is emitted when defaults are empty;. More...

Additional Inherited Members
Static Public Member Functions inherited from DefaultParamHandler
static void	writeParametersToMetaValues (const Param &write_this, MetaInfoInterface &write_here, const String &key_prefix="")
	Writes all parameters to meta values. More...

Detailed Description

Simple interface to support vector machines for classification and regression (via LIBSVM).

This class supports:

(multi-class) classification and regression with an epsilon-SVR.
linear or RBF kernel.

It uses cross-validation to optimize the respective SVM/SVR parameters C, p (SVR-only) and (RBF kernel only) gamma.

Usage: SVM models are generated by the the setup() method. SimpleSVM provides two common use cases for convinience:

all data: Passing all data (training + test) as predictors to setup and training on a subset.
training only: Passing exclusivly training data as predictors to setup. The parameter outcomes of setup() defines in both cases the training set; it contains the indexes of observations (corresponding to positions in the vectors in predictors) together with the class labels (or regression values) for training.

Given N observations of M predictors, the data are coded as a map of predictors (size M), each a numeric vector of values for different observations (size N).

To predict class labels (or regression values) based on a model, use one of the predict() methods:

if setup with all data: the parameter indexes of predict() takes a vector of indexes corresponding to the observations for which predictions should be made. (With an empty vector, the default, predictions are made for all observations, including those used for training.)
if setup with training data only: by passing a new PredictorMap.

Parameters of this class are:

Name	Type	Default	Restrictions	Description
kernel	string	RBF	RBF, linear	SVM kernel
xval	int	5	min: 1	Number of partitions for cross-validation (parameter optimization)
log2_C	float list	[-5.0, -3.0, -1.0, 1.0, 3.0, 5.0, 7.0, 9.0, 11.0, 13.0, 15.0]		Values to try for the SVM parameter 'C' during parameter optimization. A value 'x' is used as 'C = 2^x'.
log2_gamma	float list	[-15.0, -13.0, -11.0, -9.0, -7.0, -5.0, -3.0, -1.0, 1.0, 3.0]		Values to try for the SVM parameter 'gamma' during parameter optimization (RBF kernel only). A value 'x' is used as 'gamma = 2^x'.
log2_p	float list	[-15.0, -12.0, -9.0, -6.0, -3.32192809489, 0.0, 3.32192809489, 6.0, 9.0, 12.0, 15.0]		Values to try for the SVM parameter 'epsilon' during parameter optimization (epsilon-SVR only). A value 'x' is used as 'epsilon = 2^x'.
epsilon	float	1.0e-03	min: 0.0	Stopping criterion
cache_size	float	100.0	min: 1.0	Size of the kernel cache (in MB)
no_shrinking	string	false	true, false	Disable the shrinking heuristics

Note:

If a section name is documented, the documentation is displayed as tooltip.
Advanced parameter names are italic.

Class Documentation

◆ OpenMS::SimpleSVM::Prediction

struct OpenMS::SimpleSVM::Prediction

SVM/SVR prediction result.

Collaboration diagram for SimpleSVM::Prediction:

[legend]

Class Members
double	outcome	Predicted class label (or regression value)
map< double, double >	probabilities	Class label (or regression value) and their predicted probabilities.

Member Typedef Documentation

◆ PredictorMap

typedef std::map<String, std::vector<double> > PredictorMap

Mapping from predictor name to vector of predictor values.

◆ ScaleMap

typedef std::map<String, std::pair<double, double> > ScaleMap

Mapping from predictor name to predictor min and max.

◆ SVMPerformance

typedef std::vector<std::vector<std::vector<double> > > SVMPerformance

protected

Classification (or regression) performance for different param. combinations (C/gamma/p):

Constructor & Destructor Documentation

◆ SimpleSVM()

SimpleSVM ( )

Default constructor.

◆ ~SimpleSVM()

~SimpleSVM ( )

override

Destructor.

Member Function Documentation

◆ chooseBestParameters_()

std::tuple<double, double, double> chooseBestParameters_ ( bool higher_better ) const

protected

Choose best SVM parameters based on cross-validation results.

◆ clear_()

void clear_ ( )

protected

◆ convertData_()

void convertData_ ( const PredictorMap & predictors )

protected

Convert predictors to LIBSVM format.

◆ getFeatureWeights()

void getFeatureWeights ( std::map< String, double > & feature_weights ) const

Get the weights used for features (predictors) in the SVM model.

Currently only supported for two-class classification. If a linear kernel is used, the weights are informative for ranking features.

Exceptions

Exception::Precondition if no model has been trained, or if the classification involves more than two classes

◆ getScaling()

const ScaleMap& getScaling ( ) const

Get data range of predictors before scaling to [0, 1].

◆ optimizeParameters_()

void optimizeParameters_ ( bool classification )

protected

Run cross-validation to optimize SVM parameters.

◆ predict() [1/2]

void predict	(	PredictorMap &	predictors,
		std::vector< Prediction > &	predictions
	)		const

Predict class labels or regression values (and probabilities).

Parameters

predictors	Mapping from predictor name to vector of predictor values (for different observations). All vectors should have the same length; values will be changed by scaling applied to training data in setup.
predictions	Output vector of prediction results (same order as `indexes`).

Exceptions

Exception::Precondition	if no model has been trained
Exception::InvalidValue	if an invalid index is used in `indexes`

◆ predict() [2/2]

void predict	(	std::vector< Prediction > &	predictions,
		std::vector< Size >	indexes = `std::vector< Size >()`
	)		const

Predict class labels or regression values (and probabilities).

Parameters

predictions	Output vector of prediction results (same order as `indexes`).
indexes	Vector of observation indexes for which predictions are desired. If empty (default), predictions are made for all observations.

Exceptions

Exception::Precondition	if no model has been trained
Exception::InvalidValue	if an invalid index is used in `indexes`

◆ printNull_()

static void printNull_ ( const char * )

inlinestaticprotected

Dummy function to suppress LIBSVM output.

◆ scaleData_()

void scaleData_ ( PredictorMap & predictors )

protected

Scale predictor values to range 0-1.

◆ setup()

void setup	(	PredictorMap &	predictors,
		const std::map< Size, double > &	outcomes,
		bool	classification = `true`
	)

Load data and train a model.

Parameters

predictors	Mapping from predictor name to vector of predictor values (for different observations). All vectors should have the same length; values will be changed by scaling.
outcomes	Mapping from observation index to class label or regression value in the training set.
classification	true (default) if SVM classification should be used, SVR otherwise

Exceptions

Exception::IllegalArgument	if `predictors` is empty
Exception::InvalidValue	if an invalid index is used in `outcomes`
Exception::MissingInformation	if there are fewer than two class labels in `outcomes`, or if there are not enough observations for cross-validation

◆ writeXvalResults()

void writeXvalResults ( const String & path ) const

Write cross-validation (parameter optimization) results to a CSV file.

Member Data Documentation

◆ data_

struct svm_problem data_

protected

SVM training data (LIBSVM format)

◆ log2_C_

std::vector<double> log2_C_

protected

Parameter values to try during optimization.

◆ log2_gamma_

std::vector<double> log2_gamma_

protected

◆ log2_p_

std::vector<double> log2_p_

protected

◆ model_

struct svm_model* model_

protected

Pointer to SVM model (LIBSVM format)

◆ n_parts_

Size n_parts_

protected

Number of partitions for cross-validation.

◆ nodes_

std::vector<std::vector<struct svm_node> > nodes_

protected

Values of predictors (LIBSVM format)

◆ performance_

SVMPerformance performance_

protected

Cross-validation results.

◆ predictor_names_

std::vector<String> predictor_names_

protected

Names of predictors in the model (excluding uninformative ones)

◆ scaling_

ScaleMap scaling_

protected

Mapping from predictor name to predictor min and max.

◆ svm_params_

struct svm_parameter svm_params_

protected

SVM parameters (LIBSVM format)

Classes

Public Types

Public Member Functions

Protected Types

Protected Member Functions

Static Protected Member Functions

Protected Attributes

Additional Inherited Members

Detailed Description

Class Documentation

◆ OpenMS::SimpleSVM::Prediction

Member Typedef Documentation

◆ PredictorMap

◆ ScaleMap

◆ SVMPerformance

Constructor & Destructor Documentation

◆ SimpleSVM()

◆ ~SimpleSVM()

Member Function Documentation

◆ chooseBestParameters_()

◆ clear_()

◆ convertData_()

◆ getFeatureWeights()

◆ getScaling()

◆ optimizeParameters_()

◆ predict() [1/2]

◆ predict() [2/2]

◆ printNull_()

◆ scaleData_()

◆ setup()

◆ writeXvalResults()

Member Data Documentation

◆ data_

◆ log2_C_

◆ log2_gamma_

◆ log2_p_

◆ model_

◆ n_parts_

◆ nodes_

◆ performance_

◆ predictor_names_

◆ scaling_

◆ svm_params_