GBDT algorithm implementation. including Training, prediction, bagging. More...

#include <gbdt.h>

Inheritance diagram for LightGBM::GBDT:

Public Member Functions
	GBDT ()
	Constructor.

	~GBDT ()
	Destructor.

void	Init (const Config gbdt_config, const Dataset train_data, const ObjectiveFunction objective_function, const std::vector< const Metric > &training_metrics) override
	Initialization logic.

void	MergeFrom (const Boosting *other) override
	Merge model from other boosting object. Will insert to the front of current boosting object.

void	ShuffleModels (int start_iter, int end_iter) override
	Shuffle Existing Models.

void	ResetTrainingData (const Dataset train_data, const ObjectiveFunction objective_function, const std::vector< const Metric * > &training_metrics) override
	Reset the training data.

void	ResetConfig (const Config *gbdt_config) override
	Reset Boosting Config.

void	AddValidDataset (const Dataset valid_data, const std::vector< const Metric > &valid_metrics) override
	Adding a validation dataset.

void	Train (int snapshot_freq, const std::string &model_output_path) override
	Perform a full training procedure.

void	RefitTree (const std::vector< std::vector< int > > &tree_leaf_prediction) override
	Update the tree output by new training data.

virtual bool	TrainOneIter (const score_t gradients, const score_t hessians) override
	Training logic.

void	RollbackOneIter () override
	Rollback one iteration.

int	GetCurrentIteration () const override
	Get current iteration.

bool	NeedAccuratePrediction () const override
	Can use early stopping for prediction or not.

std::vector< double >	GetEvalAt (int data_idx) const override
	Get evaluation result at data_idx data.

virtual const double *	GetTrainingScore (int64_t *out_len) override
	Get current training score.

virtual int64_t	GetNumPredictAt (int data_idx) const override
	Get size of prediction at data_idx data.

void	GetPredictAt (int data_idx, double out_result, int64_t out_len) override
	Get prediction result at data_idx data.

int	NumPredictOneRow (int num_iteration, bool is_pred_leaf, bool is_pred_contrib) const override
	Get number of prediction for one data.

void	PredictRaw (const double features, double output, const PredictionEarlyStopInstance *earlyStop) const override
	Prediction for one record, not sigmoid transform.

void	PredictRawByMap (const std::unordered_map< int, double > &features, double output, const PredictionEarlyStopInstance early_stop) const override

void	Predict (const double features, double output, const PredictionEarlyStopInstance *earlyStop) const override
	Prediction for one record, sigmoid transformation will be used if needed.

void	PredictByMap (const std::unordered_map< int, double > &features, double output, const PredictionEarlyStopInstance early_stop) const override

void	PredictLeafIndex (const double features, double output) const override
	Prediction for one record with leaf index.

void	PredictLeafIndexByMap (const std::unordered_map< int, double > &features, double *output) const override

void	PredictContrib (const double features, double output, const PredictionEarlyStopInstance *earlyStop) const override
	Feature contributions for the model's prediction of one record.

std::string	DumpModel (int start_iteration, int num_iteration) const override
	Dump model to json format string.

std::string	ModelToIfElse (int num_iteration) const override
	Translate model to if-else statement.

bool	SaveModelToIfElse (int num_iteration, const char *filename) const override
	Translate model to if-else statement.

virtual bool	SaveModelToFile (int start_iteration, int num_iterations, const char *filename) const override
	Save model to file.

std::string	SaveModelToString (int num_iterations)
	Save model to string.

virtual std::string	SaveModelToString (int start_iteration, int num_iterations) const override

bool	LoadModelFromString (std::string str)
	Restore from a serialized buffer.

bool	LoadModelFromString (const char *buffer, size_t len) override

std::vector< double >	FeatureImportance (int num_iteration, int importance_type) const override
	Calculate feature importances.

int	MaxFeatureIdx () const override
	Get max feature index of this model.

std::vector< std::string >	FeatureNames () const override
	Get feature names of this model.

int	LabelIdx () const override
	Get index of label column.

int	NumberOfTotalModel () const override
	Get number of weak sub-models.

int	NumModelPerIteration () const override
	Get number of tree per iteration.

int	NumberOfClasses () const override
	Get number of classes.

void	InitPredict (int num_iteration, bool is_pred_contrib) override
	Initial work for the prediction.

double	GetLeafValue (int tree_idx, int leaf_idx) const override

void	SetLeafValue (int tree_idx, int leaf_idx, double val) override

virtual const char *	SubModelName () const override
	Get Type name of this boosting object.

Public Member Functions inherited from LightGBM::Boosting
virtual	~Boosting ()
	virtual destructor

std::string	SaveModelToString (int num_iterations)
	Save model to string.

bool	LoadModelFromString (std::string str)
	Restore from a serialized string.

Boosting &	operator= (const Boosting &)=delete
	Disable copy.

	Boosting (const Boosting &)=delete
	Disable copy.

Protected Member Functions
virtual bool	EvalAndCheckEarlyStopping ()
	Print eval result and check early stopping.

void	ResetBaggingConfig (const Config *config, bool is_change_dataset)
	reset config for bagging

virtual void	Bagging (int iter)
	Implement bagging logic.

data_size_t	BaggingHelper (Random &cur_rand, data_size_t start, data_size_t cnt, data_size_t *buffer)
	Helper function for bagging, used for multi-threading optimization.

virtual void	Boosting ()
	calculate the object function

virtual void	UpdateScore (const Tree *tree, const int cur_tree_id)
	updating score after tree was trained

virtual std::vector< double >	EvalOneMetric (const Metric metric, const double score) const
	eval results for one metric

std::string	OutputMetric (int iter)
	Print metric result of current iteration.

double	BoostFromAverage (int class_id, bool update_scorer)

Protected Attributes
int	iter_
	current iteration

const Dataset *	train_data_
	Pointer to training data.

std::unique_ptr< Config >	config_
	Config of gbdt.

std::unique_ptr< TreeLearner >	tree_learner_
	Tree learner, will use this class to learn trees.

const ObjectiveFunction *	objective_function_
	Objective function.

std::unique_ptr< ScoreUpdater >	train_score_updater_
	Store and update training data's score.

std::vector< const Metric * >	training_metrics_
	Metrics for training data.

std::vector< std::unique_ptr< ScoreUpdater > >	valid_score_updater_
	Store and update validation data's scores.

std::vector< std::vector< const Metric * > >	valid_metrics_
	Metric for validation data.

int	early_stopping_round_
	Number of rounds for early stopping.

std::vector< std::vector< int > >	best_iter_
	Best iteration(s) for early stopping.

std::vector< std::vector< double > >	best_score_
	Best score(s) for early stopping.

std::vector< std::vector< std::string > >	best_msg_
	output message of best iteration

std::vector< std::unique_ptr< Tree > >	models_
	Trained models(trees)

int	max_feature_idx_
	Max feature index of training data.

std::vector< score_t >	gradients_
	First order derivative of training data.

std::vector< score_t >	hessians_
	Secend order derivative of training data.

std::vector< data_size_t >	bag_data_indices_
	Store the indices of in-bag data.

data_size_t	bag_data_cnt_
	Number of in-bag data.

std::vector< data_size_t >	tmp_indices_
	Store the indices of in-bag data.

data_size_t	num_data_
	Number of training data.

int	num_tree_per_iteration_
	Number of trees per iterations.

int	num_class_
	Number of class.

data_size_t	label_idx_
	Index of label column.

int	num_iteration_for_pred_
	number of used model

double	shrinkage_rate_
	Shrinkage rate for one iteration.

int	num_init_iteration_
	Number of loaded initial models.

std::vector< std::string >	feature_names_
	Feature names.

std::vector< std::string >	feature_infos_

int	num_threads_
	number of threads

std::vector< data_size_t >	offsets_buf_
	Buffer for multi-threading bagging.

std::vector< data_size_t >	left_cnts_buf_
	Buffer for multi-threading bagging.

std::vector< data_size_t >	right_cnts_buf_
	Buffer for multi-threading bagging.

std::vector< data_size_t >	left_write_pos_buf_
	Buffer for multi-threading bagging.

std::vector< data_size_t >	right_write_pos_buf_
	Buffer for multi-threading bagging.

std::unique_ptr< Dataset >	tmp_subset_

bool	is_use_subset_

std::vector< bool >	class_need_train_

bool	is_constant_hessian_

std::unique_ptr< ObjectiveFunction >	loaded_objective_

bool	average_output_

bool	need_re_bagging_

std::string	loaded_parameter_

Json	forced_splits_json_

Additional Inherited Members
Static Public Member Functions inherited from LightGBM::Boosting
static bool	LoadFileToBoosting (Boosting boosting, const char filename)

static Boosting *	CreateBoosting (const std::string &type, const char *filename)
	Create boosting object.

Detailed Description

GBDT algorithm implementation. including Training, prediction, bagging.

Member Function Documentation

◆ AddValidDataset()

void LightGBM::GBDT::AddValidDataset	(	const Dataset *	valid_data,
		const std::vector< const Metric * > &	valid_metrics
	)

overridevirtual

Adding a validation dataset.

Parameters

valid_data	Validation dataset
valid_metrics	Metrics for validation dataset

Implements LightGBM::Boosting.

Reimplemented in LightGBM::RF.

◆ Bagging()

void LightGBM::GBDT::Bagging ( int iter )

protectedvirtual

Implement bagging logic.

Parameters

iter	Current interation

Reimplemented in LightGBM::GOSS.

◆ BaggingHelper()

data_size_t LightGBM::GBDT::BaggingHelper	(	Random &	cur_rand,
		data_size_t	start,
		data_size_t	cnt,
		data_size_t *	buffer
	)

protected

Helper function for bagging, used for multi-threading optimization.

Parameters

start	start indice of bagging
cnt	count
buffer	output buffer

Returns: count of left size

◆ Boosting()

void LightGBM::GBDT::Boosting ( )

protectedvirtual

calculate the object function

Reimplemented in LightGBM::RF.

◆ DumpModel()

std::string LightGBM::GBDT::DumpModel	(	int	start_iteration,
		int	num_iteration
	)		const

overridevirtual

Dump model to json format string.

Parameters

start_iteration	The model will be saved start from
num_iteration	Number of iterations that want to dump, -1 means dump all

Returns: Json format string of model

Implements LightGBM::Boosting.

◆ EvalAndCheckEarlyStopping()

bool LightGBM::GBDT::EvalAndCheckEarlyStopping ( )

protectedvirtual

Print eval result and check early stopping.

Reimplemented in LightGBM::DART.

◆ FeatureImportance()

std::vector< double > LightGBM::GBDT::FeatureImportance	(	int	num_iteration,
		int	importance_type
	)		const

overridevirtual

Calculate feature importances.

Parameters

num_iteration	Number of model that want to use for feature importance, -1 means use all
importance_type	0 for split, 1 for gain

Returns: vector of feature_importance

Implements LightGBM::Boosting.

◆ FeatureNames()

std::vector< std::string > LightGBM::GBDT::FeatureNames ( ) const

inlineoverridevirtual

Get feature names of this model.

Returns: Feature names of this model

Implements LightGBM::Boosting.

◆ GetCurrentIteration()

int LightGBM::GBDT::GetCurrentIteration ( ) const

inlineoverridevirtual

Get current iteration.

Implements LightGBM::Boosting.

◆ GetEvalAt()

std::vector< double > LightGBM::GBDT::GetEvalAt ( int data_idx ) const

overridevirtual

Get evaluation result at data_idx data.

Get eval result.

Parameters

data_idx 0: training data, 1: 1st validation data

Returns: evaluation result

Implements LightGBM::Boosting.

◆ GetLeafValue()

double LightGBM::GBDT::GetLeafValue	(	int	tree_idx,
		int	leaf_idx
	)		const

inlineoverridevirtual

Implements LightGBM::GBDTBase.

◆ GetNumPredictAt()

virtual int64_t LightGBM::GBDT::GetNumPredictAt ( int data_idx ) const

inlineoverridevirtual

Get size of prediction at data_idx data.

Parameters

data_idx 0: training data, 1: 1st validation data

Returns: The size of prediction

Implements LightGBM::Boosting.

◆ GetPredictAt()

void LightGBM::GBDT::GetPredictAt	(	int	data_idx,
		double *	out_result,
		int64_t *	out_len
	)

overridevirtual

Get prediction result at data_idx data.

Parameters

data_idx	0: training data, 1: 1st validation data
result	used to store prediction result, should allocate memory before call this function
out_len	length of returned score

Implements LightGBM::Boosting.

◆ GetTrainingScore()

const double * LightGBM::GBDT::GetTrainingScore ( int64_t * out_len )

overridevirtual

Get current training score.

Get training scores result.

Parameters

out_len length of returned score

Returns: training score

Implements LightGBM::Boosting.

Reimplemented in LightGBM::DART.

◆ Init()

void LightGBM::GBDT::Init	(	const Config *	gbdt_config,
		const Dataset *	train_data,
		const ObjectiveFunction *	objective_function,
		const std::vector< const Metric * > &	training_metrics
	)

overridevirtual

Initialization logic.

Parameters

gbdt_config	Config for boosting
train_data	Training data
objective_function	Training objective function
training_metrics	Training metrics

Implements LightGBM::Boosting.

Reimplemented in LightGBM::GOSS, and LightGBM::RF.

◆ InitPredict()

void LightGBM::GBDT::InitPredict	(	int	num_iteration,
		bool	is_pred_contrib
	)

inlineoverridevirtual

Initial work for the prediction.

Parameters

num_iteration	number of used iteration
is_pred_contrib

Implements LightGBM::Boosting.

◆ LabelIdx()

int LightGBM::GBDT::LabelIdx ( ) const

inlineoverridevirtual

Get index of label column.

Returns: index of label column

Implements LightGBM::Boosting.

◆ LoadModelFromString()

bool LightGBM::GBDT::LoadModelFromString	(	const char *	buffer,
		size_t	len
	)

overridevirtual

Implements LightGBM::Boosting.

◆ MaxFeatureIdx()

int LightGBM::GBDT::MaxFeatureIdx ( ) const

inlineoverridevirtual

Get max feature index of this model.

Returns: Max feature index of this model

Implements LightGBM::Boosting.

◆ MergeFrom()

void LightGBM::GBDT::MergeFrom ( const Boosting * other )

inlineoverridevirtual

Merge model from other boosting object. Will insert to the front of current boosting object.

Parameters

other

Implements LightGBM::Boosting.

◆ ModelToIfElse()

std::string LightGBM::GBDT::ModelToIfElse ( int num_iteration ) const

overridevirtual

Translate model to if-else statement.

Parameters

num_iteration Number of iterations that want to translate, -1 means translate all

Returns: if-else format codes of model

Implements LightGBM::Boosting.

◆ NeedAccuratePrediction()

bool LightGBM::GBDT::NeedAccuratePrediction ( ) const

inlineoverridevirtual

Can use early stopping for prediction or not.

Returns: True if cannot use early stopping for prediction

Implements LightGBM::Boosting.

Reimplemented in LightGBM::RF.

◆ NumberOfClasses()

int LightGBM::GBDT::NumberOfClasses ( ) const

inlineoverridevirtual

Get number of classes.

Returns: Number of classes

Implements LightGBM::Boosting.

◆ NumberOfTotalModel()

int LightGBM::GBDT::NumberOfTotalModel ( ) const

inlineoverridevirtual

Get number of weak sub-models.

Returns: Number of weak sub-models

Implements LightGBM::Boosting.

◆ NumModelPerIteration()

int LightGBM::GBDT::NumModelPerIteration ( ) const

inlineoverridevirtual

Get number of tree per iteration.

Returns: number of tree per iteration

Implements LightGBM::Boosting.

◆ NumPredictOneRow()

int LightGBM::GBDT::NumPredictOneRow	(	int	num_iteration,
		bool	is_pred_leaf,
		bool	is_pred_contrib
	)		const

inlineoverridevirtual

Get number of prediction for one data.

Parameters

num_iteration	number of used iterations
is_pred_leaf	True if predicting leaf index
is_pred_contrib	True if predicting feature contribution

Returns: number of prediction

Implements LightGBM::Boosting.

◆ OutputMetric()

std::string LightGBM::GBDT::OutputMetric ( int iter )

protected

Print metric result of current iteration.

Parameters

iter	Current interation

Returns: best_msg if met early_stopping

◆ Predict()

void LightGBM::GBDT::Predict	(	const double *	features,
		double *	output,
		const PredictionEarlyStopInstance *	early_stop
	)		const

overridevirtual

Prediction for one record, sigmoid transformation will be used if needed.

Parameters

feature_values	Feature value on this record
output	Prediction result for this record
early_stop	Early stopping instance. If nullptr, no early stopping is applied and all models are evaluated.

Implements LightGBM::Boosting.

◆ PredictByMap()

void LightGBM::GBDT::PredictByMap	(	const std::unordered_map< int, double > &	features,
		double *	output,
		const PredictionEarlyStopInstance *	early_stop
	)		const

overridevirtual

Implements LightGBM::Boosting.

◆ PredictContrib()

void LightGBM::GBDT::PredictContrib	(	const double *	features,
		double *	output,
		const PredictionEarlyStopInstance *	early_stop
	)		const

overridevirtual

Feature contributions for the model's prediction of one record.

Parameters

feature_values	Feature value on this record
output	Prediction result for this record
early_stop	Early stopping instance. If nullptr, no early stopping is applied and all models are evaluated.

Implements LightGBM::Boosting.

◆ PredictLeafIndex()

void LightGBM::GBDT::PredictLeafIndex	(	const double *	features,
		double *	output
	)		const

overridevirtual

Prediction for one record with leaf index.

Parameters

feature_values	Feature value on this record
output	Prediction result for this record

Implements LightGBM::Boosting.

◆ PredictLeafIndexByMap()

void LightGBM::GBDT::PredictLeafIndexByMap	(	const std::unordered_map< int, double > &	features,
		double *	output
	)		const

overridevirtual

Implements LightGBM::Boosting.

◆ PredictRaw()

void LightGBM::GBDT::PredictRaw	(	const double *	features,
		double *	output,
		const PredictionEarlyStopInstance *	early_stop
	)		const

overridevirtual

Prediction for one record, not sigmoid transform.

Parameters

feature_values	Feature value on this record
output	Prediction result for this record
early_stop	Early stopping instance. If nullptr, no early stopping is applied and all models are evaluated.

Implements LightGBM::Boosting.

◆ PredictRawByMap()

void LightGBM::GBDT::PredictRawByMap	(	const std::unordered_map< int, double > &	features,
		double *	output,
		const PredictionEarlyStopInstance *	early_stop
	)		const

overridevirtual

Implements LightGBM::Boosting.

◆ RefitTree()

void LightGBM::GBDT::RefitTree ( const std::vector< std::vector< int > > & tree_leaf_prediction )

overridevirtual

Update the tree output by new training data.

Implements LightGBM::Boosting.

◆ ResetConfig()

void LightGBM::GBDT::ResetConfig ( const Config * gbdt_config )

overridevirtual

Reset Boosting Config.

Parameters

gbdt_config Config for boosting

Implements LightGBM::Boosting.

Reimplemented in LightGBM::GOSS, and LightGBM::RF.

◆ ResetTrainingData()

void LightGBM::GBDT::ResetTrainingData	(	const Dataset *	train_data,
		const ObjectiveFunction *	objective_function,
		const std::vector< const Metric * > &	training_metrics
	)

overridevirtual

Reset the training data.

Parameters

train_data	New Training data
objective_function	Training objective function
training_metrics	Training metrics

Implements LightGBM::Boosting.

Reimplemented in LightGBM::GOSS, and LightGBM::RF.

◆ RollbackOneIter()

void LightGBM::GBDT::RollbackOneIter ( )

overridevirtual

Rollback one iteration.

Implements LightGBM::Boosting.

Reimplemented in LightGBM::RF.

◆ SaveModelToFile()

bool LightGBM::GBDT::SaveModelToFile	(	int	start_iteration,
		int	num_iterations,
		const char *	filename
	)		const

overridevirtual

Save model to file.

Parameters

start_iteration	The model will be saved start from
num_iterations	Number of model that want to save, -1 means save all
filename	Filename that want to save to

Returns: is_finish Is training finished or not

File to write models

Implements LightGBM::Boosting.

◆ SaveModelToIfElse()

bool LightGBM::GBDT::SaveModelToIfElse	(	int	num_iteration,
		const char *	filename
	)		const

overridevirtual

Translate model to if-else statement.

Parameters

num_iteration	Number of iterations that want to translate, -1 means translate all
filename	Filename that want to save to

Returns: is_finish Is training finished or not

File to write models

Implements LightGBM::Boosting.

◆ SaveModelToString() [1/2]

std::string LightGBM::GBDT::SaveModelToString ( int num_iterations )

inline

Save model to string.

Parameters

start_iteration	The model will be saved start from
num_iterations	Number of model that want to save, -1 means save all

Returns: Non-empty string if succeeded

◆ SaveModelToString() [2/2]

std::string LightGBM::GBDT::SaveModelToString	(	int	start_iteration,
		int	num_iterations
	)		const

overridevirtual

Implements LightGBM::Boosting.

◆ SetLeafValue()

void LightGBM::GBDT::SetLeafValue	(	int	tree_idx,
		int	leaf_idx,
		double	val
	)

inlineoverridevirtual

Implements LightGBM::GBDTBase.

◆ ShuffleModels()

void LightGBM::GBDT::ShuffleModels	(	int	start_iter,
		int	end_iter
	)

inlineoverridevirtual

Shuffle Existing Models.

Implements LightGBM::Boosting.

◆ SubModelName()

virtual const char * LightGBM::GBDT::SubModelName ( ) const

inlineoverridevirtual

Get Type name of this boosting object.

Implements LightGBM::Boosting.

◆ Train()

void LightGBM::GBDT::Train	(	int	snapshot_freq,
		const std::string &	model_output_path
	)

overridevirtual

Perform a full training procedure.

Parameters

snapshot_freq	frequence of snapshot
model_output_path	path of model file

Implements LightGBM::Boosting.

◆ TrainOneIter()

bool LightGBM::GBDT::TrainOneIter	(	const score_t *	gradients,
		const score_t *	hessians
	)

overridevirtual

Training logic.

Parameters

gradients	nullptr for using default objective, otherwise use self-defined boosting
hessians	nullptr for using default objective, otherwise use self-defined boosting

Returns: True if cannot train any more

Implements LightGBM::Boosting.

Reimplemented in LightGBM::DART, and LightGBM::RF.

◆ UpdateScore()

void LightGBM::GBDT::UpdateScore	(	const Tree *	tree,
		const int	cur_tree_id
	)

protectedvirtual

updating score after tree was trained

Parameters

tree	Trained tree of this iteration
cur_tree_id	Current tree for multiclass training

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

External/LightGBM_2.2.3/LightGBM-2.2.3/src/boosting/gbdt.h
External/LightGBM_2.2.3/LightGBM-2.2.3/src/boosting/gbdt.cpp
External/LightGBM_2.2.3/LightGBM-2.2.3/src/boosting/gbdt_model_text.cpp
External/LightGBM_2.2.3/LightGBM-2.2.3/src/boosting/gbdt_prediction.cpp

Public Member Functions

Protected Member Functions

Protected Attributes

Additional Inherited Members

Detailed Description

Member Function Documentation

◆ AddValidDataset()

◆ Bagging()

◆ BaggingHelper()

◆ Boosting()

◆ DumpModel()

◆ EvalAndCheckEarlyStopping()

◆ FeatureImportance()

◆ FeatureNames()

◆ GetCurrentIteration()

◆ GetEvalAt()

◆ GetLeafValue()

◆ GetNumPredictAt()

◆ GetPredictAt()

◆ GetTrainingScore()

◆ Init()

◆ InitPredict()

◆ LabelIdx()

◆ LoadModelFromString()

◆ MaxFeatureIdx()

◆ MergeFrom()

◆ ModelToIfElse()

◆ NeedAccuratePrediction()

◆ NumberOfClasses()

◆ NumberOfTotalModel()

◆ NumModelPerIteration()

◆ NumPredictOneRow()

◆ OutputMetric()

◆ Predict()

◆ PredictByMap()

◆ PredictContrib()

◆ PredictLeafIndex()

◆ PredictLeafIndexByMap()

◆ PredictRaw()

◆ PredictRawByMap()

◆ RefitTree()

◆ ResetConfig()

◆ ResetTrainingData()

◆ RollbackOneIter()

◆ SaveModelToFile()

◆ SaveModelToIfElse()

◆ SaveModelToString() [1/2]

◆ SaveModelToString() [2/2]

◆ SetLeafValue()

◆ ShuffleModels()

◆ SubModelName()

◆ Train()

◆ TrainOneIter()

◆ UpdateScore()