xgboost::linear::GreedyFeatureSelector Class Reference

Select coordinate with the greatest gradient magnitude. More...

#include <coordinate_common.h>

Inheritance diagram for xgboost::linear::GreedyFeatureSelector:

Public Member Functions
void	Setup (Context const *, const gbm::GBLinearModel &model, const std::vector< GradientPair > &, DMatrix *, float, float, int param) override
	Setting up the selector state prior to looping through features.
int	NextFeature (Context const *ctx, int, const gbm::GBLinearModel &model, int group_idx, const std::vector< GradientPair > &gpair, DMatrix *p_fmat, float alpha, float lambda) override
	Select next coordinate to update.
	FeatureSelector ()=default
Public Member Functions inherited from xgboost::linear::FeatureSelector
virtual	~FeatureSelector ()=default
	virtual destructor

Protected Attributes
bst_uint	top_k_
std::vector< bst_uint >	counter_
std::vector< std::pair< double, double > >	gpair_sums_

Additional Inherited Members
Static Public Member Functions inherited from xgboost::linear::FeatureSelector
static FeatureSelector *	Create (int choice)
	factory method

Detailed Description

Select coordinate with the greatest gradient magnitude.

Note

It has O(num_feature^2) complexity. It is fully deterministic.

It allows restricting the selection to top_k features per group with the largest magnitude of univariate weight change, by passing the top_k value through the param argument of Setup(). That would reduce the complexity to O(num_feature*top_k).

Member Function Documentation

NextFeature()

int xgboost::linear::GreedyFeatureSelector::NextFeature ( Context const *  ctx, int  iteration, const gbm::GBLinearModel &  model, int  group_idx, const std::vector< GradientPair > &  gpair, DMatrix *  p_fmat, float  alpha, float  lambda  )

inlineoverridevirtual

Select next coordinate to update.

Parameters

ctx	Booster context
iteration	The iteration in a loop through features
model	The model.
group_idx	Zero-based index of the group.
gpair	The gpair.
p_fmat	The feature matrix.
alpha	Regularisation alpha.
lambda	Regularisation lambda.

Returns

The index of the selected feature. -1 indicates none selected.

Implements xgboost::linear::FeatureSelector.

Setup()

void xgboost::linear::GreedyFeatureSelector::Setup ( Context const *  , const gbm::GBLinearModel &  , const std::vector< GradientPair > &  , DMatrix *  , float  , float  , int    )

inlineoverridevirtual

Setting up the selector state prior to looping through features.

Parameters

ctx	The booster context.
model	The model.
gpair	The gpair.
p_fmat	The feature matrix.
alpha	Regularisation alpha.
lambda	Regularisation lambda.
param	A parameter with algorithm-dependent use.

Reimplemented from xgboost::linear::FeatureSelector.

---

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

• External/xgboost/src/linear/coordinate_common.h