Builder for linear 2-state classifier. More...

#include <clsfy_binary_hyperplane_gmrho_builder.h>

Inheritance diagram for clsfy_binary_hyperplane_gmrho_builder:

Public Member Functions
	clsfy_binary_hyperplane_gmrho_builder ()
double	build (clsfy_classifier_base &classifier, mbl_data_wrapper< vnl_vector< double > > &inputs, const vcl_vector< unsigned > &outputs) const
	Build a linear classifier, with the given data.
virtual double	build (clsfy_classifier_base &model, mbl_data_wrapper< vnl_vector< double > > &inputs, unsigned nClasses, const vcl_vector< unsigned > &outputs) const
	Build model from data.
short	version_no () const
	Version number for I/O.
vcl_string	is_a () const
	Name of the class.
virtual bool	is_class (vcl_string const &s) const
	Name of the class.
void	print_summary (vcl_ostream &os) const
	Print class to os.
virtual clsfy_builder_base *	clone () const
	Create a deep copy.
void	set_auto_estimate_sigma (bool bAuto)
	should sigma be estimate during the build or a pre-defined value used.
void	set_sigma_preset (double sigma_preset)
	use this for sigma if auto_estimate_sigma is true.
virtual void	b_write (vsl_b_ostream &) const
	Save class to binary file stream.
virtual void	b_read (vsl_b_istream &)
	Load class from binary file stream.
virtual clsfy_classifier_base *	new_classifier () const
	Create a new untrained linear classifier with binary output.
virtual void	config (vcl_istream &as)
	Initialise the parameters from a text stream.
Static Public Member Functions
static vcl_auto_ptr < clsfy_builder_base >	new_builder (vcl_istream &as)
	Load description from a text stream.
Private Member Functions
double	estimate_sigma (const vnl_matrix< double > &data, const vnl_vector< double > &y) const
	Estimate the scale (sigma) used in the Geman-McClure function.
void	determine_weights (const vnl_matrix< double > &data, const vnl_vector< double > &y, double sigma) const
	Determine the weights for the hyperplane.
Private Attributes
vnl_vector< double >	weights_
	The classifier weights (weight N is the constant).
unsigned	num_examples_
	Number of training examples (data.rows()).
unsigned	num_vars_
	Number of variables (data.cols()).
double	epsilon_
	Tolerance for non-linear optimiser convergence.
bool	auto_estimate_sigma_
	should sigma be estimated during the build or a pre-defined value used.
double	sigma_preset_
	use this for sigma if auto_estimate_sigma is true.

Detailed Description

Builder for linear 2-state classifier.

Uses a Geman-McClure robust function, rather than a least squares fit, on points that are not mis-classified This increases the weighting given to points near the boundary in determining its fit A conventional least squares fit is perform first to determine a starting solution, and also the sigma scaling factor used in the GM function. Several iterations are performed during which sigma is reduced (i.e. deterministic annealing), to try and avoid local minima

Definition at line 29 of file clsfy_binary_hyperplane_gmrho_builder.h.

Constructor & Destructor Documentation

clsfy_binary_hyperplane_gmrho_builder::clsfy_binary_hyperplane_gmrho_builder ( ) [inline]

Definition at line 58 of file clsfy_binary_hyperplane_gmrho_builder.h.

Member Function Documentation

void clsfy_binary_hyperplane_gmrho_builder::b_read ( vsl_b_istream & bfs ) [virtual]

Load class from binary file stream.

Reimplemented from clsfy_binary_hyperplane_ls_builder.

Definition at line 274 of file clsfy_binary_hyperplane_gmrho_builder.cxx.

void clsfy_binary_hyperplane_gmrho_builder::b_write ( vsl_b_ostream & bfs ) const [virtual]

Save class to binary file stream.

Reimplemented from clsfy_binary_hyperplane_ls_builder.

Definition at line 265 of file clsfy_binary_hyperplane_gmrho_builder.cxx.

double clsfy_binary_hyperplane_gmrho_builder::build	(	clsfy_classifier_base &	classifier,
		mbl_data_wrapper< vnl_vector< double > > &	inputs,
		const vcl_vector< unsigned > &	outputs
	)		const

Build a linear classifier, with the given data.

Build a linear hyperplane classifier with the given data.

Return the mean error over the training set.

Reduce the influence of well classified points far into their correct region by applying a Geman-McClure robust error function, rather than a least squares fit

Reimplemented from clsfy_binary_hyperplane_ls_builder.

Definition at line 110 of file clsfy_binary_hyperplane_gmrho_builder.cxx.

double clsfy_binary_hyperplane_gmrho_builder::build	(	clsfy_classifier_base &	classifier,
		mbl_data_wrapper< vnl_vector< double > > &	inputs,
		unsigned	n_classes,
		const vcl_vector< unsigned > &	outputs
	)		const `[virtual]`

Build model from data.

Build a linear classifier, with the given data.

Return the mean error over the training set. For this classifiers, you must nClasses==1 to indicate a binary classifier

Return the mean error over the training set. n_classes must be 1.

Reimplemented from clsfy_binary_hyperplane_ls_builder.

Definition at line 98 of file clsfy_binary_hyperplane_gmrho_builder.cxx.

clsfy_builder_base * clsfy_binary_hyperplane_gmrho_builder::clone ( ) const [virtual]

Create a deep copy.

client is responsible for deleting returned object.

Reimplemented from clsfy_binary_hyperplane_ls_builder.

Definition at line 321 of file clsfy_binary_hyperplane_gmrho_builder.cxx.

void clsfy_builder_base::config ( vcl_istream & as ) [virtual, inherited]

Initialise the parameters from a text stream.

Default case accepts no parameters.

Reimplemented in clsfy_rbf_svm_smo_1_builder, clsfy_parzen_builder, clsfy_knn_builder, clsfy_null_builder, and clsfy_binary_1d_wrapper_builder.

Definition at line 26 of file clsfy_builder_base.cxx.

void clsfy_binary_hyperplane_gmrho_builder::determine_weights	(	const vnl_matrix< double > &	data,
		const vnl_vector< double > &	y,
		double	sigma
	)		const `[private]`

Determine the weights for the hyperplane.

Definition at line 194 of file clsfy_binary_hyperplane_gmrho_builder.cxx.

double clsfy_binary_hyperplane_gmrho_builder::estimate_sigma	(	const vnl_matrix< double > &	data,
		const vnl_vector< double > &	y
	)		const `[private]`

Estimate the scale (sigma) used in the Geman-McClure function.

This is increased by the mis-classification overlap region if any

Definition at line 211 of file clsfy_binary_hyperplane_gmrho_builder.cxx.

vcl_string clsfy_binary_hyperplane_gmrho_builder::is_a ( ) const [virtual]

Name of the class.

Reimplemented from clsfy_binary_hyperplane_ls_builder.

Definition at line 294 of file clsfy_binary_hyperplane_gmrho_builder.cxx.

bool clsfy_binary_hyperplane_gmrho_builder::is_class ( vcl_string const & s ) const [virtual]

Name of the class.

Reimplemented from clsfy_binary_hyperplane_ls_builder.

Definition at line 301 of file clsfy_binary_hyperplane_gmrho_builder.cxx.

vcl_auto_ptr< clsfy_builder_base > clsfy_builder_base::new_builder ( vcl_istream & as ) [static, inherited]

Load description from a text stream.

The stream should contain the name of the feature extractor class that will be used, followed by a brace-enclosed list of parameters for the builder. This function will construct the appropriate clsfy_builder_base derivative and return that.

Exceptions: