BALL::QSAR::Model Class Reference

#include <BALL/QSAR/Model.h>

Inheritance diagram for BALL::QSAR::Model:

Public Member Functions
Constructors and Destructors
	Model (const QSARData &q)
virtual	~Model ()
virtual void	operator= (const Model &m)

Protected Member Functions
Input and Output. The following methods can be used to implement the functions saveToFile() and readFromFile() in final classes derived from this base-class
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	saveDescriptorInformationToFile (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)

Protected Attributes
int	default_no_opt_steps_

Friends
class	Validation
class	RegressionValidation
class	ClassificationValidation
class	PCRModel
class	KPCRModel
class	FeatureSelection

Accessors
void	copyData (const Model &m)
void	copyDescriptorIDs (const Model &m)
void	readTrainingData ()
virtual Eigen::VectorXd	predict (const vector< double > &substance, bool transform)=0
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)
virtual void	saveToFile (string filename)=0
virtual void	readFromFile (string filename)=0
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)
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 ()

Attributes
const QSARData *	data
Validation *	model_val
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 34 of file Model.h.

Constructor & Destructor Documentation

BALL::QSAR::Model::Model

(

const QSARData &

q

)

constructur

Parameters

q	QSARData object, from which the data for this model should be taken

virtual BALL::QSAR::Model::~Model ( )

virtual

Member Function Documentation

void BALL::QSAR::Model::addLambda ( Eigen::MatrixXd &  matrix, double &  lambda  )

protected

adds offset lambda to the diagonal of the given matrix

void BALL::QSAR::Model::backTransformPrediction ( Eigen::VectorXd &  pred )

protected

transforms a prediction (obtained by Model.train()) according to the inverse of the transformation(s) of the activity values of the training data

virtual double BALL::QSAR::Model::calculateStdErr ( )

inlinevirtual

Reimplemented in BALL::QSAR::GPModel.

Definition at line 93 of file Model.h.

void BALL::QSAR::Model::copyData

(

const Model &

m

)

copies the data (descriptor matrix, names of substances and descriptors) and the IDs of the selected descriptors from m

void BALL::QSAR::Model::copyDescriptorIDs

(

const Model &

m

)

copies the IDs of the selected descriptors from m

void BALL::QSAR::Model::deleteDescriptorIDs

(

)

removes all entries from descriptor_IDs

std::multiset<unsigned int>* BALL::QSAR::Model::getDescriptorIDs

(

)

returns a const pointer to the descriptor IDs of this model

const Eigen::MatrixXd* BALL::QSAR::Model::getDescriptorMatrix

(

)

returns a const pointer to the descriptor matrix of this model

const vector<string>* BALL::QSAR::Model::getDescriptorNames

(

)

returns a const pointer to the names of the descriptors of this model

const Eigen::MatrixXd BALL::QSAR::Model::getDescriptorTransformations

(

)

returns descriptor transformations

virtual vector<double> BALL::QSAR::Model::getParameters ( ) const

virtual

Reimplemented in BALL::QSAR::KPLSModel, BALL::QSAR::PLSModel, BALL::QSAR::KPCRModel, BALL::QSAR::GPModel, BALL::QSAR::PCRModel, BALL::QSAR::NBModel, BALL::QSAR::OPLSModel, BALL::QSAR::ALLModel, BALL::QSAR::SNBModel, BALL::QSAR::RRModel, BALL::QSAR::LDAModel, BALL::QSAR::LibsvmModel, and BALL::QSAR::KNNModel.

const vector<string>* BALL::QSAR::Model::getSubstanceNames

(

)

returns a const pointer to the names of the substances of this model

Eigen::VectorXd BALL::QSAR::Model::getSubstanceVector ( const vector< double > &  substance, bool  transform  )

protected

returns a Row-Vector containing only the values for these descriptors, that have been selected for this model

Parameters

substance

a vector of all descriptor values for the substance to be predicted

Eigen::VectorXd BALL::QSAR::Model::getSubstanceVector ( const Eigen::VectorXd &  substance, bool  transform  )

protected

const string* BALL::QSAR::Model::getType

(

)

returns the type of the current model, e.g. "MLR", "PLS", ...

void BALL::QSAR::Model::getUnnormalizedFeatureValue	(	int	compound,
		int	feature,
		double &	return_value
	)

Fetches the un-normalized value for the specified feature of the desired compound (instance) from the data that this Model currently contains. This method is needed for visualization purposes only.

void BALL::QSAR::Model::getUnnormalizedResponseValue	(	int	compound,
		int	response,
		double &	return_value
	)

Fetches the un-normalized value for the specified response of the desired compound (instance) from the data that this Model currently contains. This method is needed for visualization purposes only.

const Eigen::MatrixXd* BALL::QSAR::Model::getY

(

)

returns a const pointer to the activity values of this model

const Eigen::MatrixXd BALL::QSAR::Model::getYTransformations

(

)

virtual void BALL::QSAR::Model::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 in BALL::QSAR::KernelModel.

virtual bool BALL::QSAR::Model::optimizeParameters ( int  , int    )

inlinevirtual

optimizes parameters (!=number of features) of the model, e.g. no of latente variables in case of PLS model or kernel width in case of automated lazy learning model.
The number of selected features (=descriptors) is NOT changed by this method. Use class FeatureSelection in order to do this.

Returns

1 if parameters were optimized using cross-validation. The best Q2 value is assumed to be saved in ModelValidation.Q2
0 if the model has no parameters to be optimized, so that no cross-validation was done.

Reimplemented in BALL::QSAR::KPLSModel, BALL::QSAR::PLSModel, BALL::QSAR::OPLSModel, BALL::QSAR::ALLModel, and BALL::QSAR::KNNModel.

Definition at line 89 of file Model.h.

bool BALL::QSAR::Model::optimizeParameters

(

int

k

)

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

pure 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

Implemented in BALL::QSAR::KernelModel, BALL::QSAR::GPModel, BALL::QSAR::NBModel, BALL::QSAR::SNBModel, BALL::QSAR::LDAModel, BALL::QSAR::LinearModel, BALL::QSAR::LogitModel, and BALL::QSAR::ALLModel.

void BALL::QSAR::Model::readDescriptorInformation ( )

protected

reads selected descriptors, their names and the information about their transformations (mean and stddev of each descriptor). This function is used after feature selection to read information about the selected features

virtual void BALL::QSAR::Model::readDescriptorInformationFromFile ( std::ifstream &  in, int  no_descriptors, bool  transformation  )

protectedvirtual

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

pure virtual

reconstruct a saved Model from a file

Implemented in BALL::QSAR::ALLModel, BALL::QSAR::SNBModel, BALL::QSAR::NBModel, BALL::QSAR::RegressionModel, BALL::QSAR::LDAModel, BALL::QSAR::KernelModel, and BALL::QSAR::LogitModel.

void BALL::QSAR::Model::readMatrix ( Eigen::MatrixXd &  mat, std::ifstream &  in, unsigned int  lines, unsigned int  col  )

protected

reconstructs a Eigen::MatrixXd from a given input stream after resizing the given Eigen::MatrixXd as specified

void BALL::QSAR::Model::readModelParametersFromFile ( std::ifstream &  in )

protected

void BALL::QSAR::Model::readResponseTransformationFromFile ( std::ifstream &  in, int  no_y  )

protected

void BALL::QSAR::Model::readTrainingData

(

)

copies the data for the relevant descriptors from the bound QSARData object into this model and updates Model.descriptor_transformations and Model.y_transformations .
If no explicit feature selection was done, i.e. if descriptor_IDs is emtpy, all data is fetched.
If feature selection was done, i.e. if descriptor_IDs is not empty, only the columns of the relevant descriptors are fetched.

void BALL::QSAR::Model::readVector ( Eigen::RowVectorXd &  vec, std::ifstream &  in, unsigned int  no_cells, bool  column_vector  )

protected

virtual void BALL::QSAR::Model::saveDescriptorInformationToFile ( std::ofstream &  out )

protectedvirtual

overloaded by class RegressionModel, whose member function can also save coefficients and coefficient-errors

Reimplemented in BALL::QSAR::RegressionModel.

void BALL::QSAR::Model::saveModelParametersToFile ( std::ofstream &  out )

protected

void BALL::QSAR::Model::saveResponseTransformationToFile ( std::ofstream &  out )

protected

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

pure virtual

save Model to a file

Implemented in BALL::QSAR::ALLModel, BALL::QSAR::SNBModel, BALL::QSAR::NBModel, BALL::QSAR::RegressionModel, BALL::QSAR::LDAModel, BALL::QSAR::KernelModel, and BALL::QSAR::LogitModel.

void BALL::QSAR::Model::setDataSource

(

const QSARData *

q

)

void BALL::QSAR::Model::setDescriptorIDs

(

const std::multiset< unsigned int > &

sl

)

manually specify a set of descriptors

virtual void BALL::QSAR::Model::setParameters ( vector< double > &  )

inlinevirtual

sets the model parameters according to the given values.

Reimplemented in BALL::QSAR::KPLSModel, BALL::QSAR::PLSModel, BALL::QSAR::NBModel, BALL::QSAR::KPCRModel, BALL::QSAR::GPModel, BALL::QSAR::PCRModel, BALL::QSAR::OPLSModel, BALL::QSAR::ALLModel, BALL::QSAR::SNBModel, BALL::QSAR::RRModel, BALL::QSAR::LDAModel, BALL::QSAR::LibsvmModel, and BALL::QSAR::KNNModel.

Definition at line 97 of file Model.h.

virtual void BALL::QSAR::Model::train ( )

pure virtual

Starts training the model.

Implemented in BALL::QSAR::KPCRModel, BALL::QSAR::KPLSModel, BALL::QSAR::PLSModel, BALL::QSAR::PCRModel, BALL::QSAR::GPModel, BALL::QSAR::NBModel, BALL::QSAR::MLRModel, BALL::QSAR::SNBModel, BALL::QSAR::RRModel, BALL::QSAR::ALLModel, BALL::QSAR::OPLSModel, BALL::QSAR::LDAModel, BALL::QSAR::LibsvmModel, and BALL::QSAR::LogitModel.

Friends And Related Function Documentation

friend class ClassificationValidation

friend

Definition at line 237 of file Model.h.

friend class FeatureSelection

friend

Definition at line 240 of file Model.h.

friend class KPCRModel

friend

Definition at line 239 of file Model.h.

friend class PCRModel

friend

Definition at line 238 of file Model.h.

friend class RegressionValidation

friend

Definition at line 236 of file Model.h.

friend class Validation

friend

Definition at line 235 of file Model.h.

Member Data Documentation

const QSARData* BALL::QSAR::Model::data

pointer to the input data class for this model

Definition at line 147 of file Model.h.

int BALL::QSAR::Model::default_no_opt_steps_

protected

The default number of steps for model parameter optimization.
It can be adjusted by the different types of models.
Standard default value is 30.

Definition at line 159 of file Model.h.

std::multiset<unsigned int> BALL::QSAR::Model::descriptor_IDs_

protected

list containing the IDs of the selected descriptors (=features); with IDs >= 0
If this list is empty, it is assumed that no feature selection was done, i.e. that all descriptors are to be considered for cross-validation and prediction of activity.
If it is not empty, only the descriptors in this list are used for cross-validation and prediction of activity.

Definition at line 232 of file Model.h.

Eigen::MatrixXd BALL::QSAR::Model::descriptor_matrix_

protected

matrix containing the values of each descriptor for each substance

Definition at line 206 of file Model.h.

vector<string> BALL::QSAR::Model::descriptor_names_

protected

names of all descriptors

Definition at line 212 of file Model.h.

Eigen::MatrixXd BALL::QSAR::Model::descriptor_transformations_

protected

2xm dimensional matrix (m=no of descriptors) containing mean and stddev of each selected descriptor.
The content of this matrix is updated only by Model.readTrainingData()

Definition at line 216 of file Model.h.

Validation* BALL::QSAR::Model::model_val

a ModelValidation object, that is used to validate this model and that will contain the results of the validations

Definition at line 150 of file Model.h.

vector<string> BALL::QSAR::Model::substance_names_

protected

names of all substances

Definition at line 209 of file Model.h.

String BALL::QSAR::Model::type_

protected

The type of model, e.g. "MLR", "GP", ...

Definition at line 227 of file Model.h.

Eigen::MatrixXd BALL::QSAR::Model::Y_

protected

Matrix containing the experimentally determined results (active/non-active) for each substance.
Each column contains the values for one activity.

Definition at line 224 of file Model.h.

Eigen::MatrixXd BALL::QSAR::Model::y_transformations_

protected

2xc dimensional matrix (c=no of activities) containing mean and stddev of each activity.
The content of this matrix is updated only by Model.readTrainingData()

Definition at line 220 of file Model.h.