Train SVM models that are used by SvmTheoreticalSpectrumGenerator. More...

#include <OpenMS/CHEMISTRY/SvmTheoreticalSpectrumGeneratorTrainer.h>

Inheritance diagram for SvmTheoreticalSpectrumGeneratorTrainer:

Public Member Functions
SvmTheoreticalSpectrumGeneratorTrainer &	operator= (const SvmTheoreticalSpectrumGeneratorTrainer &tsg)
	Assignment operator.
void	trainModel (const PeakMap &spectra, const std::vector< AASequence > &annotations, String filename, Int precursor_charge)
	trains an SVM for each ion_type and stores them in files <filename>_residue_loss_charge.svm
void	normalizeIntensity (PeakSpectrum &S) const
	Normalizes the intensity of the peaks in the input data.
Constructors and Destructors

	SvmTheoreticalSpectrumGeneratorTrainer ()
	Default constructor.
	SvmTheoreticalSpectrumGeneratorTrainer (const SvmTheoreticalSpectrumGeneratorTrainer &source)
	Copy constructor.
virtual	~SvmTheoreticalSpectrumGeneratorTrainer ()
	Destructor.
Protected Member Functions
void	writeTrainingFile_ (std::vector< DescriptorSet > &training_input, std::vector< DoubleReal > &training_output, String filename)
	Write a training file that can be passed to libsvm command line tools.
Private Types
typedef SvmTheoreticalSpectrumGenerator::IonType	IonType
typedef SvmTheoreticalSpectrumGenerator::DescriptorSet	DescriptorSet
typedef std::map< std::pair < IonType, Size >, std::vector < DoubleReal > >	ObservedIntensMap
Private Member Functions
void	countIntensities_ (const PeakSpectrum &spectrum, const AASequence &annotation, IonType type, std::map< std::pair< IonType, Size >, std::vector< DoubleReal > > &observed_intensities, DoubleReal tolerance, Size number_of_regions)
	stores the observed intensities for each sector-type combination in a vector
void	trainSecondaryTypes_ (TextFile &info_outfile, Size number_of_regions, Size number_of_intensity_levels, ObservedIntensMap &observed_intensities, const std::vector< IonType > &ion_types, const std::vector< bool > &is_primary)
	trains the Bayesian secondary peak types models

Detailed Description

Train SVM models that are used by SvmTheoreticalSpectrumGenerator.

Parameters of this class are:

Name	Type	Default	Restrictions	Description
write_training_files	string	false	true, false	If set to true no models are trained but files (__training.dat) are produced for the selected primary ion types. They can be used as input for LibSVM command line tools
number_intensity_levels	int	7		The number of intensity bins (for secondary type models)
number_regions	int	3		The number of regions each spectrum is split to (for secondary type models)
parent_tolerance	float	2.5		The maximum difference between theoretical and experimental parent mass to accept training spectrum
peak_tolerance	float	0.5		The maximum mass error for a peak to the expected mass of some ion type
add_b_ions	string	true	true, false	Train simulator for b-ions
add_y_ions	string	true	true, false	Train simulator for y-ions
add_a_ions	string	false	true, false	Train simulator for a-ions
add_c_ions	string	false	true, false	Train simulator for c-ions
add_x_ions	string	false	true, false	Train simulator for x-ions
add_z_ions	string	false	true, false	Train simulator for z-ions
add_losses	string	false	true, false	Train simulator for neutral losses of H2O and NH3 for b-ions and y-ions
add_b2_ions	string	false	true, false	Train simulator for doubly charged b-ions
add_y2_ions	string	false	true, false	Train simulator for double charged y-ions
svm:svc_type	int	0	min: 0 max: 1	Type of the SVC: 0=C_SVC 1=NU_SVC
svm:svr_type	int	1	min: 0 max: 1	Type of the SVR: 0=EPSILON_SVR 1=NU_SVR
svm:scaling	string	true	true, false	Apply scaling of feature values
svm:scaling_lower	float	0		Lower bound for scaling
svm:scaling_upper	float	1		Upper bound for scaling
svm:n_fold	int	5	min: 1	n_fold cross validation is performed
svm:grid	string	false	true, false	Perform grid search
svm:additive_cv	string	false	true, false	Additive step size (if false multiplicative)
svm:svc:kernel_type	int	2	min: 0 max: 3	Type of the kernel: 0=LINEAR 1=POLY 2=RBF 3=SIGMOID
svm:svc:degree	int	3	min: 1	For POLY
svm:svc:gamma	float	0	min: 0	For POLY/RBF/SIGMOID
svm:svc:C	float	1		Cost of constraint violation
svm:svc:nu	float	0.5		For NU_SVC, ONE_CLASS and NU_SVR
svm:svc:balancing	string	true	true, false	Use class balanced SVC training
svm:svc:degree_start	int	1	min: 1	starting point of degree
svm:svc:degree_step_size	int	2		step size point of degree
svm:svc:degree_stop	int	4		stopping point of degree
svm:svc:gamma_start	float	1e-05	min: 0 max: 1	starting point of gamma
svm:svc:gamma_step_size	int	100		step size point of gamma
svm:svc:gamma_stop	float	0.1		stopping point of gamma
svm:svc:c_start	float	0.1		starting point of c
svm:svc:c_step_size	int	100		step size of c
svm:svc:c_stop	int	1000		stopping point of c
svm:svc:nu_start	float	0.3	min: 0 max: 1	starting point of nu
svm:svc:nu_step_size	int	2		step size of nu
svm:svc:nu_stop	float	0.6	min: 0 max: 1	stopping point of nu
svm:svr:kernel_type	int	2	min: 0 max: 3	Type of the kernel: 0=LINEAR 1=POLY 2=RBF 3=SIGMOID
svm:svr:degree	int	3	min: 1	For POLY
svm:svr:gamma	float	0	min: 0	For POLY/RBF/SIGMOID
svm:svr:C	float	1		Cost of constraint violation
svm:svr:p	float	0.1		The epsilon for the loss function in epsilon-SVR
svm:svr:nu	float	0.5		For NU_SVC, ONE_CLASS and NU_SVR
svm:svr:degree_start	int	1	min: 1	starting point of degree
svm:svr:degree_step_size	int	2		step size point of degree
svm:svr:degree_stop	int	4		stopping point of degree
svm:svr:gamma_start	float	1e-05	min: 0 max: 1	starting point of gamma
svm:svr:gamma_step_size	int	100		step size point of gamma
svm:svr:gamma_stop	float	0.1		stopping point of gamma
svm:svr:p_start	float	1e-05		starting point of p
svm:svr:p_step_size	int	100		step size point of p
svm:svr:p_stop	float	0.1		stopping point of p
svm:svr:c_start	float	0.1		starting point of c
svm:svr:c_step_size	int	100		step size of c
svm:svr:c_stop	int	1000		stopping point of c
svm:svr:nu_start	float	0.3	min: 0 max: 1	starting point of nu
svm:svr:nu_step_size	int	2		step size of nu
svm:svr:nu_stop	float	0.6	min: 0 max: 1	stopping point of nu

Note:

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

This class implements the algorithm used by the homonymous tool which can be used to train models for MS/MS spectrum simulation.
For the primary ion types (y, b) a SVM is trained using the libSVM library.
All important libSVM parameters are accessible as parameters.
Please refer to the libSVM manuals for detailed description of the parameters. Default values are choses as in the svm-training tool delivered with libSVM.

For the secondary types (a, c, x, z, losses, b2, y2) a simple Bayesian model is used.

Member Typedef Documentation

typedef SvmTheoreticalSpectrumGenerator::DescriptorSet DescriptorSet [private]

typedef SvmTheoreticalSpectrumGenerator::IonType IonType [private]

typedef std::map<std::pair<IonType, Size>, std::vector<DoubleReal> > ObservedIntensMap [private]

Constructor & Destructor Documentation

SvmTheoreticalSpectrumGeneratorTrainer ( )

Default constructor.

SvmTheoreticalSpectrumGeneratorTrainer ( const SvmTheoreticalSpectrumGeneratorTrainer & source )

Copy constructor.

virtual ~SvmTheoreticalSpectrumGeneratorTrainer ( ) [virtual]

Destructor.

Member Function Documentation

void countIntensities_	(	const PeakSpectrum &	spectrum,
		const AASequence &	annotation,
		IonType	type,
		std::map< std::pair< IonType, Size >, std::vector< DoubleReal > > &	observed_intensities,
		DoubleReal	tolerance,
		Size	number_of_regions
	)			`[private]`

stores the observed intensities for each sector-type combination in a vector

void normalizeIntensity ( PeakSpectrum & S ) const

Normalizes the intensity of the peaks in the input data.

SvmTheoreticalSpectrumGeneratorTrainer& operator= ( const SvmTheoreticalSpectrumGeneratorTrainer & tsg )

Assignment operator.

void trainModel	(	const PeakMap &	spectra,
		const std::vector< AASequence > &	annotations,
		String	filename,
		Int	precursor_charge
	)

trains an SVM for each ion_type and stores them in files <filename>_residue_loss_charge.svm

void trainSecondaryTypes_	(	TextFile &	info_outfile,
		Size	number_of_regions,
		Size	number_of_intensity_levels,
		ObservedIntensMap &	observed_intensities,
		const std::vector< IonType > &	ion_types,
		const std::vector< bool > &	is_primary
	)			`[private]`

trains the Bayesian secondary peak types models

void writeTrainingFile_	(	std::vector< DescriptorSet > &	training_input,
		std::vector< DoubleReal > &	training_output,
		String	filename
	)			`[protected]`

Write a training file that can be passed to libsvm command line tools.

OpenMS / TOPP release 1.10.0

Documentation generated on Thu Mar 7 2013 09:42:50 using doxygen 1.7.1

SvmTheoreticalSpectrumGeneratorTrainer Class Reference [Chemistry]

Public Member Functions

Protected Member Functions

Private Types

Private Member Functions

Detailed Description

Member Typedef Documentation

Constructor & Destructor Documentation

Member Function Documentation

SvmTheoreticalSpectrumGeneratorTrainer Class Reference
[Chemistry]