xgboost::linear Namespace Reference

Copyright 2018-2023 by XGBoost Contributors. More...

Data Structures
struct	CoordinateParam
class	CoordinateUpdater
	Coordinate descent algorithm that updates one feature per iteration. More...
class	CyclicFeatureSelector
	Deterministic selection by cycling through features one at a time. More...
class	FeatureSelector
	Abstract class for stateful feature selection or ordering in coordinate descent algorithms. More...
class	GreedyFeatureSelector
	Select coordinate with the greatest gradient magnitude. More...
struct	LinearTrainParam
class	RandomFeatureSelector
	A random (with replacement) coordinate selector. More...
class	ShotgunUpdater
class	ShuffleFeatureSelector
	Similar to Cyclic but with random feature shuffling prior to each update. More...
class	ThriftyFeatureSelector
	Thrifty, approximately-greedy feature selector. More...

Enumerations
enum	FeatureSelectorEnum {   kCyclic = 0 , kShuffle , kThrifty , kGreedy ,   kRandom }
	A set of available FeatureSelector's.

Functions
double	CoordinateDelta (double sum_grad, double sum_hess, double w, double reg_alpha, double reg_lambda)
	Calculate change in weight for a given feature.
double	CoordinateDeltaBias (double sum_grad, double sum_hess)
	Calculate update to bias.
std::pair< double, double >	GetGradient (Context const *ctx, int group_idx, int num_group, bst_feature_t fidx, std::vector< GradientPair > const &gpair, DMatrix *p_fmat)
	Get the gradient with respect to a single feature.
std::pair< double, double >	GetGradientParallel (Context const *ctx, int group_idx, int num_group, int fidx, const std::vector< GradientPair > &gpair, DMatrix *p_fmat)
	Get the gradient with respect to a single feature.
std::pair< double, double >	GetBiasGradientParallel (int group_idx, int num_group, const std::vector< GradientPair > &gpair, DMatrix *p_fmat, int32_t n_threads)
	Get the gradient with respect to the bias.
void	UpdateResidualParallel (Context const *ctx, bst_feature_t fidx, int group_idx, int num_group, float dw, std::vector< GradientPair > *in_gpair, DMatrix *p_fmat)
	Updates the gradient vector with respect to a change in weight.
void	UpdateBiasResidualParallel (Context const *ctx, int group_idx, int num_group, float dbias, std::vector< GradientPair > *in_gpair, DMatrix *p_fmat)
	Updates the gradient vector based on a change in the bias.
	DMLC_REGISTER_PARAMETER (LinearTrainParam)
	DMLC_REGISTRY_LINK_TAG (updater_shotgun)
	DMLC_REGISTRY_LINK_TAG (updater_coordinate)
	DMLC_REGISTER_PARAMETER (CoordinateParam)
	DMLC_REGISTRY_FILE_TAG (updater_coordinate)
	describe ("Update linear model according to coordinate descent algorithm.") .set_body([]()
	DMLC_REGISTRY_FILE_TAG (updater_shotgun)
	describe ("Update linear model according to shotgun coordinate descent " "algorithm.") .set_body([]()

Detailed Description

Copyright 2018-2023 by XGBoost Contributors.

Author

Rory Mitchell

Function Documentation

CoordinateDelta()

double xgboost::linear::CoordinateDelta ( double  sum_grad, double  sum_hess, double  w, double  reg_alpha, double  reg_lambda  )

inline

Calculate change in weight for a given feature.

Applies l1/l2 penalty normalised by the number of training instances.

Parameters

sum_grad	The sum gradient.
sum_hess	The sum hess.
w	The weight.
reg_alpha	Unnormalised L1 penalty.
reg_lambda	Unnormalised L2 penalty.

Returns

The weight update.

CoordinateDeltaBias()

double xgboost::linear::CoordinateDeltaBias ( double  sum_grad, double  sum_hess  )

inline

Calculate update to bias.

Parameters

sum_grad	The sum gradient.
sum_hess	The sum hess.

Returns

The weight update.

GetBiasGradientParallel()

std::pair< double, double > xgboost::linear::GetBiasGradientParallel ( int  group_idx, int  num_group, const std::vector< GradientPair > &  gpair, DMatrix *  p_fmat, int32_t  n_threads  )

inline

Get the gradient with respect to the bias.

Row-wise multithreaded.

Parameters

group_idx	Zero-based index of the group.
num_group	Number of groups.
gpair	Gradients.
p_fmat	The feature matrix.

Returns

The gradient and diagonal Hessian entry for the bias.

GetGradient()

std::pair< double, double > xgboost::linear::GetGradient ( Context const *  ctx, int  group_idx, int  num_group, bst_feature_t  fidx, std::vector< GradientPair > const &  gpair, DMatrix *  p_fmat  )

inline

Get the gradient with respect to a single feature.

Parameters

group_idx	Zero-based index of the group.
num_group	Number of groups.
fidx	The target feature.
gpair	Gradients.
p_fmat	The feature matrix.

Returns

The gradient and diagonal Hessian entry for a given feature.

GetGradientParallel()

std::pair< double, double > xgboost::linear::GetGradientParallel ( Context const *  ctx, int  group_idx, int  num_group, int  fidx, const std::vector< GradientPair > &  gpair, DMatrix *  p_fmat  )

inline

Get the gradient with respect to a single feature.

Row-wise multithreaded.

Parameters

group_idx	Zero-based index of the group.
num_group	Number of groups.
fidx	The target feature.
gpair	Gradients.
p_fmat	The feature matrix.

Returns

The gradient and diagonal Hessian entry for a given feature.

UpdateBiasResidualParallel()

void xgboost::linear::UpdateBiasResidualParallel ( Context const *  ctx, int  group_idx, int  num_group, float  dbias, std::vector< GradientPair > *  in_gpair, DMatrix *  p_fmat  )

inline

Updates the gradient vector based on a change in the bias.

Parameters

group_idx	Zero-based index of the group.
num_group	Number of groups.
dbias	The change in bias.
in_gpair	The gradient vector to be updated.
p_fmat	The input feature matrix.

UpdateResidualParallel()

void xgboost::linear::UpdateResidualParallel ( Context const *  ctx, bst_feature_t  fidx, int  group_idx, int  num_group, float  dw, std::vector< GradientPair > *  in_gpair, DMatrix *  p_fmat  )

inline

Updates the gradient vector with respect to a change in weight.

Parameters

fidx	The feature index.
group_idx	Zero-based index of the group.
num_group	Number of groups.
dw	The change in weight.
in_gpair	The gradient vector to be updated.
p_fmat	The input feature matrix.