Column major matrix for gradient index on CPU. More...

#include <column_matrix.h>

Public Member Functions
bst_feature_t	GetNumFeature () const

	ColumnMatrix (GHistIndexMatrix const &gmat, double sparse_threshold)

void	InitFromSparse (SparsePage const &page, const GHistIndexMatrix &gmat, double sparse_threshold, int32_t n_threads)
	Initialize ColumnMatrix from GHistIndexMatrix with reference to the original SparsePage.

void	InitFromGHist (Context const *ctx, GHistIndexMatrix const &gmat)
	Initialize ColumnMatrix from GHistIndexMatrix without reference to actual data.

bool	IsInitialized () const

template<typename Batch >
void	PushBatch (int32_t n_threads, Batch const &batch, float missing, GHistIndexMatrix const &gmat, size_t base_rowid)
	Push batch of data for Quantile DMatrix support.

void	SetTypeSize (size_t max_bin_per_feat)

template<typename BinIdxType >
auto	SparseColumn (bst_feature_t fidx, bst_row_t first_row_idx) const

template<typename BinIdxType , bool any_missing>
auto	DenseColumn (bst_feature_t fidx) const

template<typename RowBinIdxT >
void	SetIndexNoMissing (bst_row_t base_rowid, RowBinIdxT const *row_index, const size_t n_samples, const size_t n_features, int32_t n_threads)

template<typename Batch >
void	SetIndexMixedColumns (size_t base_rowid, Batch const &batch, const GHistIndexMatrix &gmat, float missing)
	Set column index for both dense and sparse columns.

void	SetIndexMixedColumns (const GHistIndexMatrix &gmat)
	Set column index for both dense and sparse columns, but with only GHistMatrix available and requires a search for each bin.

BinTypeSize	GetTypeSize () const

auto	GetColumnType (bst_feature_t fidx) const

bool	AnyMissing () const

bool	Read (AlignedResourceReadStream fi, uint32_t const index_base)

std::size_t	Write (AlignedFileWriteStream *fo) const

MissingIndicator const &	Missing () const

Detailed Description

Column major matrix for gradient index on CPU.

This matrix contains both dense columns and sparse columns, the type of the column is controlled by the sparse threshold parameter. When the number of missing values in a column is below the threshold it's classified as dense column.

Member Function Documentation

◆ InitFromGHist()

void xgboost::common::ColumnMatrix::InitFromGHist	(	Context const *	ctx,
		GHistIndexMatrix const &	gmat
	)

inline

Initialize ColumnMatrix from GHistIndexMatrix without reference to actual data.

This function requires a binary search for each bin to get back the feature index for those bins.

◆ PushBatch()

template<typename Batch >

void xgboost::common::ColumnMatrix::PushBatch	(	int32_t	n_threads,
		Batch const &	batch,
		float	missing,
		GHistIndexMatrix const &	gmat,
		size_t	base_rowid
	)

inline

Push batch of data for Quantile DMatrix support.

Parameters

batch	Input data wrapped inside a adapter batch.
gmat	The row-major histogram index that contains index for ALL data.
base_rowid	The beginning row index for current batch.

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

External/xgboost/src/common/column_matrix.h
External/xgboost/src/common/column_matrix.cc

Public Member Functions

Detailed Description

Member Function Documentation

◆ InitFromGHist()

◆ PushBatch()