#include <lambdamart.h>

Inheritance diagram for quickrank::learning::forests::LambdaMart:

Public Member Functions
	LambdaMart (size_t ntrees, double shrinkage, size_t nthresholds, size_t ntreeleaves, size_t minleafsupport, size_t esr)
	Initializes a new LambdaMart instance with the given learning parameters. More...

	LambdaMart (const pugi::xml_document &model)
	Generates a LTR_Algorithm instance from a previously saved XML model. More...

virtual	~LambdaMart ()

virtual std::string	name () const
	Returns the name of the ranker. More...

Public Member Functions inherited from quickrank::learning::forests::Mart
	Mart (size_t ntrees, double shrinkage, size_t nthresholds, size_t ntreeleaves, size_t minleafsupport, size_t valid_iterations)
	Initializes a new Mart instance with the given learning parameters. More...

	Mart (const pugi::xml_document &model)
	Generates a LTR_Algorithm instance from a previously saved XML model. More...

virtual	~Mart ()

virtual void	learn (std::shared_ptr< data::Dataset > training_dataset, std::shared_ptr< data::Dataset > validation_dataset, std::shared_ptr< metric::ir::Metric > training_metric, size_t partial_save, const std::string output_basename)
	Start the learning process. More...

virtual Score	score_document (const Feature *d) const
	Returns the score by the current ranker. More...

virtual std::shared_ptr< std::vector< Score > >	partial_scores_document (const Feature *d) const
	Returns the partial scores of a given document, tree. More...

virtual void	print_additional_stats (void) const
	Print additional statistics. More...

virtual bool	update_weights (std::shared_ptr< std::vector< double >> weights)
	Update the weights for the ensemble models (only). More...

virtual std::shared_ptr< std::vector< double > >	get_weights () const
	Return the weights for the ensemble models (only). More...

Public Member Functions inherited from quickrank::learning::LTR_Algorithm
	LTR_Algorithm ()

	LTR_Algorithm (const pugi::xml_document &model)
	Generates a LTR_Algorithm instance from a previously saved XML model. More...

virtual	~LTR_Algorithm ()

	LTR_Algorithm (const LTR_Algorithm &other)=delete
	Avoid inefficient copy constructor. More...

LTR_Algorithm &	operator= (const LTR_Algorithm &)=delete
	Avoid inefficient copy assignment. More...

virtual void	score_dataset (std::shared_ptr< data::Dataset > dataset, Score *scores) const
	Given and input dateset, the current ranker generates scores for each instance and store the in the scores vector. More...

virtual void	save (std::string model_filename, int suffix=-1) const
	Save the current model to the output_file. More...

Static Public Attributes
static const std::string	NAME_ = "LAMBDAMART"

Static Public Attributes inherited from quickrank::learning::forests::Mart
static const std::string	NAME_ = "MART"

Protected Member Functions
virtual void	init (std::shared_ptr< data::VerticalDataset > training_dataset)
	Prepares private data structurs befor training takes place. More...

virtual void	clear (size_t num_features)
	De-allocates private data structure after training has taken place. More...

virtual void	compute_pseudoresponses (std::shared_ptr< data::VerticalDataset > training_dataset, metric::ir::Metric *metric)
	Computes pseudo responses. More...

virtual std::unique_ptr< RegressionTree >	fit_regressor_on_gradient (std::shared_ptr< data::VerticalDataset > training_dataset)
	Fits a regression tree on the gradient given by the pseudo residuals. More...

Protected Member Functions inherited from quickrank::learning::forests::Mart
virtual void	update_modelscores (std::shared_ptr< data::Dataset > dataset, Score scores, RegressionTree tree)
	Updates scores with the last learnt regression tree. More...

virtual void	update_modelscores (std::shared_ptr< data::VerticalDataset > dataset, Score scores, RegressionTree tree)

virtual pugi::xml_document *	get_xml_model () const
	Return the xml model representing the current object. More...

Protected Attributes
double *	instance_weights_ = NULL

Protected Attributes inherited from quickrank::learning::forests::Mart
float **	thresholds_ = NULL

size_t *	thresholds_size_ = NULL

double *	scores_on_training_ = NULL

quickrank::Score *	scores_on_validation_ = NULL

size_t	validation_bestmodel_ = 0

double *	pseudoresponses_ = NULL

Ensemble	ensemble_model_

size_t	ntrees_

double	shrinkage_

size_t	nthresholds_

size_t	nleaves_

size_t	minleafsupport_

size_t	valid_iterations_

size_t **	sortedsid_ = NULL

size_t	sortedsize_ = 0

RTRootHistogram *	hist_ = NULL

Additional Inherited Members
Static Public Member Functions inherited from quickrank::learning::LTR_Algorithm
static std::shared_ptr< LTR_Algorithm >	load_model_from_file (std::string model_filename)
	Load a model from a given XML file. More...

Constructor & Destructor Documentation

quickrank::learning::forests::LambdaMart::LambdaMart	(	size_t	ntrees,
		double	shrinkage,
		size_t	nthresholds,
		size_t	ntreeleaves,
		size_t	minleafsupport,
		size_t	esr
	)

inline

Initializes a new LambdaMart instance with the given learning parameters.

Parameters

ntrees	Maximum number of trees.
shrinkage	Learning rate.
nthresholds	Number of bins in discretization. 0 means no discretization.
ntreeleaves	Maximum number of leaves in each tree.
minleafsupport	Minimum number of instances in each leaf.
esr	Early stopping if no improvement after iterations on the validation set.

quickrank::learning::forests::LambdaMart::LambdaMart ( const pugi::xml_document & model )

inline

Generates a LTR_Algorithm instance from a previously saved XML model.

virtual quickrank::learning::forests::LambdaMart::~LambdaMart ( )

inlinevirtual

Member Function Documentation

void quickrank::learning::forests::LambdaMart::clear ( size_t num_features )

protectedvirtual

De-allocates private data structure after training has taken place.

Reimplemented from quickrank::learning::forests::Mart.

void quickrank::learning::forests::LambdaMart::compute_pseudoresponses	(	std::shared_ptr< data::VerticalDataset >	training_dataset,
		metric::ir::Metric *	metric
	)

protectedvirtual

Computes pseudo responses.

Parameters

training_dataset	The training data.
metric	The metric to be optimized.

Reimplemented from quickrank::learning::forests::Mart.

std::unique_ptr< RegressionTree > quickrank::learning::forests::LambdaMart::fit_regressor_on_gradient ( std::shared_ptr< data::VerticalDataset > training_dataset )