xgboost::MetaInfo Class Reference

Meta information about dataset, always sit in memory. More...

#include <data.h>

Public Member Functions
	MetaInfo ()=default
	default constructor
	MetaInfo (MetaInfo &&that)=default
MetaInfo &	operator= (MetaInfo &&that)=default
MetaInfo &	operator= (MetaInfo const &that)=delete
void	Validate (int32_t device) const
	Validate all metainfo.
MetaInfo	Slice (common::Span< int32_t const > ridxs) const
MetaInfo	Copy () const
bst_float	GetWeight (size_t i) const
	Get weight of each instances.
const std::vector< size_t > &	LabelAbsSort (Context const *ctx) const
	get sorted indexes (argsort) of labels by absolute value (used by cox loss)
void	Clear ()
	clear all the information
void	LoadBinary (dmlc::Stream *fi)
	Load the Meta info from binary stream.
void	SaveBinary (dmlc::Stream *fo) const
	Save the Meta info to binary stream.
void	SetInfo (Context const &ctx, const char *key, const void *dptr, DataType dtype, size_t num)
	Set information in the meta info.
void	SetInfo (Context const &ctx, StringView key, StringView interface_str)
	Set information in the meta info with array interface.
void	GetInfo (char const *key, bst_ulong *out_len, DataType dtype, const void **out_dptr) const
void	SetFeatureInfo (const char *key, const char **info, const bst_ulong size)
void	GetFeatureInfo (const char *field, std::vector< std::string > *out_str_vecs) const
void	Extend (MetaInfo const &that, bool accumulate_rows, bool check_column)
void	SynchronizeNumberOfColumns ()
	Synchronize the number of columns across all workers.
bool	IsRowSplit () const
	Whether the data is split row-wise.
bool	IsColumnSplit () const
	Whether the data is split column-wise.
bool	IsRanking () const
	Whether this is a learning to rank data.
bool	IsVerticalFederated () const
	A convenient method to check if we are doing vertical federated learning, which requires some special processing.
bool	ShouldHaveLabels () const
	A convenient method to check if the MetaInfo should contain labels.

Data Fields
uint64_t	num_row_ {0}
	number of rows in the data
uint64_t	num_col_ {0}
	number of columns in the data
uint64_t	num_nonzero_ {0}
	number of nonzero entries in the data
linalg::Tensor< float, 2 >	labels
	label of each instance
DataSplitMode	data_split_mode {DataSplitMode::kRow}
	data split mode
std::vector< bst_group_t >	group_ptr_
	the index of begin and end of a group needed when the learning task is ranking.
HostDeviceVector< bst_float >	weights_
	weights of each instance, optional
linalg::Tensor< float, 2 >	base_margin_
	initialized margins, if specified, xgboost will start from this init margin can be used to specify initial prediction to boost from.
HostDeviceVector< bst_float >	labels_lower_bound_
	lower bound of the label, to be used for survival analysis (censored regression)
HostDeviceVector< bst_float >	labels_upper_bound_
	upper bound of the label, to be used for survival analysis (censored regression)
std::vector< std::string >	feature_type_names
	Name of type for each feature provided by users. Eg. "int"/"float"/"i"/"q".
std::vector< std::string >	feature_names
	Name for each feature.
HostDeviceVector< FeatureType >	feature_types
HostDeviceVector< float >	feature_weights

Static Public Attributes
static constexpr uint64_t	kNumField = 12
	number of data fields in MetaInfo

Detailed Description

Meta information about dataset, always sit in memory.

Member Function Documentation

GetWeight()

bst_float xgboost::MetaInfo::GetWeight ( size_t  i ) const

inline

Get weight of each instances.

Parameters

i	Instance index.

Returns

The weight.

LoadBinary()

void xgboost::MetaInfo::LoadBinary

(

dmlc::Stream *

fi

)

Load the Meta info from binary stream.

Parameters

fi	The input stream

SaveBinary()

void xgboost::MetaInfo::SaveBinary

(

dmlc::Stream *

fo

)

const

Save the Meta info to binary stream.

Parameters

fo	The output stream.

SetInfo() [1/2]

void xgboost::MetaInfo::SetInfo	(	Context const &	ctx,
		const char *	key,
		const void *	dptr,
		DataType	dtype,
		size_t	num
	)

Set information in the meta info.

Parameters

key	The key of the information.
dptr	The data pointer of the source array.
dtype	The type of the source data.
num	Number of elements in the source array.

SetInfo() [2/2]

void xgboost::MetaInfo::SetInfo	(	Context const &	ctx,
		StringView	key,
		StringView	interface_str
	)

Set information in the meta info with array interface.

Parameters

key	The key of the information.
interface_str	String representation of json format array interface.

ShouldHaveLabels()

bool xgboost::MetaInfo::ShouldHaveLabels

(

)

const

A convenient method to check if the MetaInfo should contain labels.

Normally we assume labels are available everywhere. The only exception is in vertical federated learning where labels are only available on worker 0.

SynchronizeNumberOfColumns()

void xgboost::MetaInfo::SynchronizeNumberOfColumns

(

)

Synchronize the number of columns across all workers.

Normally we just need to find the maximum number of columns across all workers, but in vertical federated learning, since each worker loads its own list of columns, we need to sum them.

---

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

• External/xgboost/include/xgboost/data.h
• External/xgboost/src/data/data.cc