BALL::QSAR::KernelModel Class Reference

#include <BALL/QSAR/kernelModel.h>

Inheritance diagram for BALL::QSAR::KernelModel:

Public Member Functions
Constructors and Destructors
	KernelModel (const QSARData &q, int k_type, double p1, double p2)
	KernelModel (const QSARData &q, String f, String g)
	KernelModel (const QSARData &q, Eigen::VectorXd &w)
	KernelModel (const QSARData &q, const LinearModel &lm, int column)
	~KernelModel ()
virtual void	saveToFile (string filename)
virtual void	readFromFile (string filename)
virtual Eigen::VectorXd	predict (const vector< double > &substance, bool transform)
void	operator= (const Model &m)
	NonLinearModel (const QSARData &q)
	~NonLinearModel ()
Public Member Functions inherited from BALL::QSAR::RegressionModel
	RegressionModel (const QSARData &q)
	~RegressionModel ()
virtual EIGEN_MAKE_ALIGNED_OPERATOR_NEW void	operator= (const RegressionModel &m)
const Eigen::MatrixXd *	getTrainingResult () const
void	show ()
Public Member Functions inherited from BALL::QSAR::Model
	Model (const QSARData &q)
virtual	~Model ()
void	copyData (const Model &m)
void	copyDescriptorIDs (const Model &m)
void	readTrainingData ()
void	deleteDescriptorIDs ()
virtual void	train ()=0
virtual bool	optimizeParameters (int, int)
bool	optimizeParameters (int k)
virtual double	calculateStdErr ()
virtual void	setParameters (vector< double > &)
virtual vector< double >	getParameters () const
std::multiset< unsigned int > *	getDescriptorIDs ()
void	setDataSource (const QSARData *q)
const Eigen::MatrixXd *	getDescriptorMatrix ()
const vector< string > *	getSubstanceNames ()
const vector< string > *	getDescriptorNames ()
const Eigen::MatrixXd	getDescriptorTransformations ()
const Eigen::MatrixXd	getYTransformations ()
const Eigen::MatrixXd *	getY ()
void	setDescriptorIDs (const std::multiset< unsigned int > &sl)
const string *	getType ()
void	getUnnormalizedFeatureValue (int compound, int feature, double &return_value)
void	getUnnormalizedResponseValue (int compound, int response, double &return_value)

Attributes
Kernel *	kernel
Eigen::MatrixXd	K_

Input and Output. The following methods can be used to implement the functions saveToFile() and readFromFile() in final classes derived from this base-class
resulting matrix with one column for each modeled activity and with one coefficient for each substance (i.e. one column for each column for Model.Y)
class	RegressionValidation
void	calculateOffsets ()
void	readKernelParametersFromFile (std::ifstream &in)
void	saveKernelParametersToFile (std::ofstream &out)
void	saveTrainingResult (std::ofstream &out)
void	readTrainingResult (std::ifstream &input, int no_substances, int no_y)

Additional Inherited Members
RegressionValidation *	validation
const QSARData *	data
Validation *	model_val
void	readDescriptorInformationFromFile (std::ifstream &in, int no_descriptors, bool transformation, int no_coefficients)
void	saveDescriptorInformationToFile (std::ofstream &out)
Protected Member Functions inherited from BALL::QSAR::Model
void	readMatrix (Eigen::MatrixXd &mat, std::ifstream &in, unsigned int lines, unsigned int col)
void	readVector (Eigen::RowVectorXd &vec, std::ifstream &in, unsigned int no_cells, bool column_vector)
void	readModelParametersFromFile (std::ifstream &in)
void	saveModelParametersToFile (std::ofstream &out)
virtual void	readDescriptorInformationFromFile (std::ifstream &in, int no_descriptors, bool transformation)
void	readResponseTransformationFromFile (std::ifstream &in, int no_y)
void	saveResponseTransformationToFile (std::ofstream &out)
Eigen::VectorXd	getSubstanceVector (const vector< double > &substance, bool transform)
Eigen::VectorXd	getSubstanceVector (const Eigen::VectorXd &substance, bool transform)
void	backTransformPrediction (Eigen::VectorXd &pred)
void	addLambda (Eigen::MatrixXd &matrix, double &lambda)
void	readDescriptorInformation ()
Protected Attributes inherited from BALL::QSAR::RegressionModel
Eigen::MatrixXd	training_result_
Eigen::RowVectorXd	offsets_
int	default_no_opt_steps_
Eigen::MatrixXd	descriptor_matrix_
vector< string >	substance_names_
vector< string >	descriptor_names_
Eigen::MatrixXd	descriptor_transformations_
Eigen::MatrixXd	y_transformations_
Eigen::MatrixXd	Y_
String	type_
std::multiset< unsigned int >	descriptor_IDs_

Detailed Description

Definition at line 27 of file kernelModel.h.

Constructor & Destructor Documentation

KernelModel() [1/4]

BALL::QSAR::KernelModel::KernelModel	(	const QSARData &	q,
		int	k_type,
		double	p1,
		double	p2 )

KernelModel() [2/4]

BALL::QSAR::KernelModel::KernelModel	(	const QSARData &	q,
		String	f,
		String	g )

KernelModel() [3/4]

BALL::QSAR::KernelModel::KernelModel	(	const QSARData &	q,
		Eigen::VectorXd &	w )

KernelModel() [4/4]

BALL::QSAR::KernelModel::KernelModel	(	const QSARData &	q,
		const LinearModel &	lm,
		int	column )

~KernelModel()

BALL::QSAR::KernelModel::~KernelModel

(

)

Member Function Documentation

calculateOffsets()

void BALL::QSAR::KernelModel::calculateOffsets ( )

protectedvirtual

do nothing for non-kernel models. KernelModel on the other hand overloads this function...

Reimplemented from BALL::QSAR::NonLinearModel.

operator=()

void BALL::QSAR::KernelModel::operator= ( const Model & m )

virtual

copy constructur; creates a model with the same specifications as the given one (same model and kernel parameters). If the given model has been trained, the training result is copied as well.
Note, that the input data that has been read by m to m.descriptor_matrix_ and m.Y_ is NOT copied to new model, since the input data is not part of the specification of a model. If nevertheless, copying of the input data is desired, use function copyData() (afterwards).

Reimplemented from BALL::QSAR::Model.

predict()

virtual Eigen::VectorXd BALL::QSAR::KernelModel::predict ( const vector< double > & substance, bool transform )

virtual

Predicts the activities of a given substance

Parameters

substance

the substance which activity is to be predicted in form of a vecor containing the values for all descriptors (if neccessary, relevant descriptors will be selected automatically)

transform

determines whether the values for each descriptor of the given substance should be transformed before prediction of activity. If (transform==1): each descriptor value is transformed according to the centering of the respective column of QSARData.descriptor_matrix used to train this model. If the substance to be predicted is part of the same input data (e.g. same SD-file) as the training data (as is the case during cross validation), transform should therefore be set to 0.

Returns

a RowVector containing one value for each predicted activity

Implements BALL::QSAR::Model.

Reimplemented in BALL::QSAR::GPModel.

readFromFile()

virtual void BALL::QSAR::KernelModel::readFromFile ( string filename )

virtual

reconstruct a saved Model from a file

Reimplemented from BALL::QSAR::RegressionModel.

readKernelParametersFromFile()

void BALL::QSAR::KernelModel::readKernelParametersFromFile ( std::ifstream & in )

protected

readTrainingResult()

void BALL::QSAR::KernelModel::readTrainingResult ( std::ifstream & input, int no_substances, int no_y )

protected

saveKernelParametersToFile()

void BALL::QSAR::KernelModel::saveKernelParametersToFile ( std::ofstream & out )

protected

saveToFile()

virtual void BALL::QSAR::KernelModel::saveToFile ( string filename )

virtual

save Model to a file

Reimplemented from BALL::QSAR::RegressionModel.

saveTrainingResult()

void BALL::QSAR::KernelModel::saveTrainingResult ( std::ofstream & out )

protected

Friends And Related Symbol Documentation

RegressionValidation

friend class RegressionValidation

friend

Definition at line 84 of file kernelModel.h.

Member Data Documentation

K_

Eigen::MatrixXd BALL::QSAR::KernelModel::K_

protected

Matrix containing the pairwise distances between all substances

Definition at line 67 of file kernelModel.h.

kernel

Kernel* BALL::QSAR::KernelModel::kernel

Definition at line 57 of file kernelModel.h.