Implements linear regression optimization templated on a drwnFeatureMap. More...

Inheritance diagram for drwnTLinearRegressor< FeatureMap >:

Public Member Functions
	drwnTLinearRegressor ()
	default constructor

	drwnTLinearRegressor (unsigned n)
	construct a linear regressor for data of dimension `n`

	drwnTLinearRegressor (const drwnTLinearRegressor< FeatureMap > &c)
	copy constructor

virtual drwnTLinearRegressor < FeatureMap > *	clone () const
	returns a copy of the class usually implemented as virtual Foo* clone() { return new Foo(*this); }

virtual void	initialize (unsigned n)
	initialize the regressor to accept data of dimensionality `n`

virtual double	getRegression (const vector< double > &features) const
	return the estimated value for a given feature vector More...

Public Member Functions inherited from drwnLinearRegressorBase
	drwnLinearRegressorBase ()
	default constructor

	drwnLinearRegressorBase (unsigned n)
	construct a linear regressor for data of dimension `n`

	drwnLinearRegressorBase (const drwnLinearRegressorBase &r)
	copy constructor

virtual const char *	type () const
	returns object type as a string (e.g., Foo::type() { return "Foo"; })

virtual bool	save (drwnXMLNode &xml) const
	write object to XML node (see also write)

virtual bool	load (drwnXMLNode &xml)
	read object from XML node (see also read)

virtual double	train (const drwnRegressionDataset &dataset)
	estimate the regression parameters a drwnRegressionDataset

Public Member Functions inherited from drwnRegression
	drwnRegression ()
	default constructor

	drwnRegression (unsigned n)
	construct a regression object for data of dimension `n`

	drwnRegression (const drwnRegression &r)
	copy constructor

int	numFeatures () const
	return the dimensionality of the feature space

virtual bool	valid () const
	return true if the regressor has valid parameters (i.e., has been trained)

virtual double	train (const vector< vector< double > > &features, const vector< double > &targets)
	estimate the regression parameters a set of training examples

virtual double	train (const vector< vector< double > > &features, const vector< double > &targets, const vector< double > &weights)
	estimate the regression parameters a set of weighted training examples

virtual double	train (const char *filename)
	estimate the regression parameters from a drwnRegressionDataset file

virtual void	getRegressions (const vector< vector< double > > &features, vector< double > &outputTargets) const
	compute the estimated values for a set of feature vector and return the estimates in `outputTargets`

Public Member Functions inherited from drwnWriteable
bool	write (const char *filename) const
	write object to file (calls save)

bool	read (const char *filename)
	read object from file (calls load)

void	dump () const
	print object's current state to standard output (for debugging)

Public Member Functions inherited from drwnProperties
unsigned	numProperties () const

bool	hasProperty (const string &name) const

bool	hasProperty (const char *name) const

unsigned	findProperty (const string &name) const

unsigned	findProperty (const char *name) const

void	setProperty (unsigned indx, bool value)

void	setProperty (unsigned indx, int value)

void	setProperty (unsigned indx, double value)

void	setProperty (unsigned indx, const string &value)

void	setProperty (unsigned indx, const char *value)

void	setProperty (unsigned indx, const Eigen::VectorXd &value)

void	setProperty (unsigned indx, const Eigen::MatrixXd &value)

void	setProperty (const char *name, bool value)

void	setProperty (const char *name, int value)

void	setProperty (const char *name, double value)

void	setProperty (const char *name, const string &value)

void	setProperty (const char name, const char value)

void	setProperty (const char *name, const Eigen::VectorXd &value)

void	setProperty (const char *name, const Eigen::MatrixXd &value)

string	getPropertyAsString (unsigned indx) const

drwnPropertyType	getPropertyType (unsigned indx) const

bool	isReadOnly (unsigned indx) const

const drwnPropertyInterface *	getProperty (unsigned indx) const

const drwnPropertyInterface *	getProperty (const char *name) const

bool	getBoolProperty (unsigned indx) const

int	getIntProperty (unsigned indx) const

double	getDoubleProperty (unsigned indx) const

const string &	getStringProperty (unsigned indx) const

const list< string > &	getListProperty (unsigned indx) const

int	getSelectionProperty (unsigned indx) const

const Eigen::VectorXd &	getVectorProperty (unsigned indx) const

const Eigen::MatrixXd &	getMatrixProperty (unsigned indx) const

const string &	getPropertyName (unsigned indx) const

vector< string >	getPropertyNames () const

void	readProperties (drwnXMLNode &xml, const char *tag="property")

void	writeProperties (drwnXMLNode &xml, const char *tag="property") const

void	printProperties (ostream &os) const

Protected Member Functions
virtual double	objectiveAndGradient (const double x, double df) const
	returns value of objective function and populates gradient df at point x

Protected Member Functions inherited from drwnLinearRegressorBase
double	objective (const double *x) const
	returns value of objective function at point x

void	gradient (const double x, double df) const
	populates gradient of objective function at point x

Protected Member Functions inherited from drwnProperties
void	declareProperty (const string &name, drwnPropertyInterface *optif)

void	undeclareProperty (const string &name)

void	exposeProperties (drwnProperties *opts, const string &prefix=string(""), bool bSerializable=false)

virtual void	propertyChanged (const string &name)

Protected Member Functions inherited from drwnOptimizer
	drwnOptimizer ()
	default constructor

	drwnOptimizer (unsigned n)
	construct a problem with dimension n

	drwnOptimizer (const drwnOptimizer &o)
	copy constructor

void	initialize (unsigned n, const double *x=NULL)
	initialize an optimization problem of size n possibly with feasible starting point x $\mathbb{R}^n$ (or zero)

void	initialize (const double *x=NULL)
	initialize an optimization problem at feasible starting point x in $\mathbb{R}^n$ (or zero)

double	solve (unsigned maxiter, bool bMonitor=false)
	Solve the optimization problem for up to maxiter iterations to precision set by EPSF, EPSG, and EPSX static variables. Calls monitor function after each iteration if bMonitor is true.

unsigned	size () const
	dimension of optimization problem

double	operator[] (unsigned i) const
	returns the i-th component of the current solution

double &	operator[] (unsigned i)
	returns a reference to the i-th component of the current solution

virtual void	monitor (unsigned iter, double objValue)
	callback for each iteration during optimization (if bMonitor is true)

Additional Inherited Members
Static Public Attributes inherited from drwnLinearRegressorBase
static double	HUBER_BETA = 1.0e-3
	beta parameter for huber penalty

static double	REG_STRENGTH = 1.0e-9
	regularization strength

static int	MAX_ITERATIONS = 1000
	maximum training iterations

Protected Attributes inherited from drwnLinearRegressorBase
VectorXd	_theta
	regression weights

int	_penalty
	regression penalty option

double	_beta
	huber penalty threshold

int	_regularizer
	regularization option

double	_lambda
	regularization strength

const vector< vector< double > > *	_features

const vector< double > *	_targets

const vector< double > *	_weights

Protected Attributes inherited from drwnRegression
int	_nFeatures
	number of features

bool	_bValid
	true if regression parameters are trained or loaded

Protected Attributes inherited from drwnOptimizer
unsigned	_n
	dimension of optimization problem (i.e., $\mathbb{R}^n$ )

double *	_x
	current feasible solution in $\mathbb{R}^n$

double *	_df
	gradient at _x in $\mathbb{R}^n$

Static Protected Attributes inherited from drwnOptimizer
static double	EPSF = 1.0e-6
	default tolerance on function convergence

static double	EPSG = 1.0e-3
	deafult tolerance on gradient convergence

static double	EPSX = 1.0e-6
	default tolerance on solution convergence

Detailed Description

template<class FeatureMap = drwnBiasFeatureMap>
class drwnTLinearRegressor< FeatureMap >

Implements linear regression optimization templated on a drwnFeatureMap.

Parameters are learned using either L2 or Huber penalty on the prediction error and can be regularized with L2 or huber penalty.

The following code snippet shows example usage:

// load a dataset
drwnRegressionDataset dataset("training.bin");
const int numFeatures = dataset.numFeatures();
// train the regressor
drwnLinearRegressor regressor(numFeatures);
regressor.train(dataset);
// compute mean-square-error
double sse = 0.0;
for (size_t i = 0; i < dataset.size(); i++) {
    const double y = regressor.getRegression(dataset.features[i]);
    const double err = dataset.targets[i] - y;
    sse += err * err;
}
DRWN_LOG_MESSAGE("mse is " << sse / (double)dataset.size());

See Also: drwnFeatureMap, Feature Space Mappings

Member Function Documentation

template<class FeatureMap >

double drwnTLinearRegressor< FeatureMap >::getRegression ( const vector< double > & features ) const

virtual

return the estimated value for a given feature vector

Todo:: define feature mapping functions in terms of Eigen::VectorXd

Implements drwnLinearRegressorBase.

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

drwnLinearRegressor.h

Public Member Functions

Protected Member Functions

Additional Inherited Members

Detailed Description

template<class FeatureMap = drwnBiasFeatureMap> class drwnTLinearRegressor< FeatureMap >

Member Function Documentation

template<class FeatureMap = drwnBiasFeatureMap>
class drwnTLinearRegressor< FeatureMap >