xgboost Namespace Reference

Core data structure for multi-target trees. More...

Namespaces
	collective
	common
	detail
	gbm
	linalg
	system
	tree

Classes
class	GradientPairInt64
	Fixed point representation for high precision gradient pair. Has a different interface so we don't accidentally use it in gain calculations. More...
class	DMatrixCache
	Thread-aware FIFO cache for DMatrix related data. More...
struct	DeviceSym
struct	DeviceOrd
	A type for device ordinal. The type is packed into 32-bit for efficient use in viewing types like linalg::TensorView. More...
struct	Context
	Runtime context for XGBoost. Contains information like threads and device. More...
class	MetaInfo
	Meta information about dataset, always sit in memory. More...
struct	Entry
	Element from a sparse vector. More...
struct	BatchParam
	Parameters for constructing histogram index batches. More...
struct	HostSparsePageView
class	SparsePage
	In-memory storage unit of sparse batch, stored in CSR format. More...
class	CSCPage
class	ExtSparsePage
	Sparse page for exporting DMatrix. Same as SparsePage, just a different type to prevent being used internally. More...
class	SortedCSCPage
class	BatchIteratorImpl
class	BatchIterator
class	BatchSet
struct	ExtMemConfig
class	DMatrix
	Internal data structured used by XGBoost to hold all external data. More...
class	FeatureMap
	Feature map data structure to help text model dump. TODO(tqchen) consider make it even more lightweight. More...
class	GradientBooster
	interface of gradient boosting model. More...
struct	GradientBoosterReg
	Registry entry for tree updater. More...
struct	GlobalConfiguration
struct	InitNewThread
struct	HostDeviceVectorImpl
class	HostDeviceVector
class	IntrusivePtrCell
	Helper class for embedding reference counting into client objects. See https://www.boost.org/doc/libs/1_74_0/doc/html/atomic/usage_examples.html for discussions of memory order. More...
class	IntrusivePtr
	Implementation of Intrusive Pointer. A smart pointer that points to an object with an embedded reference counter. The underlying object must implement a friend function IntrusivePtrRefCount() that returns the ref counter (of type IntrusivePtrCell). The intrusive pointer is faster than std::shared_ptr<>: std::shared_ptr<> makes an extra memory allocation for the ref counter whereas the intrusive pointer does not. More...
class	Value
class	JsonString
class	JsonArray
class	JsonTypedArray
	Typed array for Universal Binary JSON. More...
class	JsonObject
class	JsonNumber
class	JsonInteger
class	JsonNull
class	JsonBoolean
	Describes both true and false. More...
class	Json
	Data structure representing JSON format. More...
class	JsonReader
	A json reader, currently error checking and utf-8 is not fully supported. More...
class	JsonWriter
class	UBJReader
	Reader for UBJSON https://ubjson.org/. More...
class	UBJWriter
	Writer for UBJSON https://ubjson.org/. More...
class	Learner
	Learner class that does training and prediction. This is the user facing module of xgboost training. The Load/Save function corresponds to the model used in python/R. More...
struct	LearnerModelParam
	Basic model parameters, used to describe the booster. More...
class	LinearUpdater
	interface of linear updater More...
struct	LinearUpdaterReg
	Registry entry for linear updater. More...
class	Metric
	interface of evaluation metric used to evaluate model performance. This has nothing to do with training, but merely act as evaluation purpose. More...
struct	MetricReg
	Registry entry for Metric factory functions. The additional parameter const char* param gives the value after @, can be null. For example, metric map@3, then: param == "3". More...
struct	Model
struct	Configurable
class	MultiTargetTree
	Tree structure for multi-target model. More...
class	ObjFunction
	interface of objective function More...
struct	ObjFunctionReg
	Registry entry for objective factory functions. More...
struct	XGBoostParameter
struct	PredictionCacheEntry
	Contains pointer to input matrix and associated cached predictions. More...
class	PredictionContainer
	A container for managed prediction caches. More...
class	Predictor
	Performs prediction on individual training instances or batches of instances for GBTree. Prediction functions all take a GBTreeModel and a DMatrix as input and output a vector of predictions. The predictor does not modify any state of the model itself. More...
struct	PredictorReg
	Registry entry for predictor. More...
struct	StringView
struct	ObjInfo
	A struct returned by objective, which determines task at hand. The struct is not used by any algorithm yet, only for future development like categorical split. More...
struct	TreeParam
	meta parameters of the tree More...
struct	RTreeNodeStat
	node statistics used in regression tree More...
class	CopyUniquePtr
	Helper for defining copyable data structure that contains unique pointers. More...
class	RegTree
	define regression tree to be the most common tree model. More...
class	TreeUpdater
	interface of tree update module, that performs update of a tree. More...
struct	TreeUpdaterReg
	Registry entry for tree updater. More...

Typedefs
using	bst_uint = std::uint32_t
	unsigned integer type used for feature index. More...
using	bst_ulong = std::uint64_t
	unsigned long integers More...
using	bst_float = float
	float type, used for storing statistics More...
using	bst_cat_t = std::int32_t
	Categorical value type. More...
using	bst_feature_t = std::uint32_t
	Type for data column (feature) index. More...
using	bst_bin_t = std::int32_t
	Type for histogram bin index. We sometimes use -1 to indicate invalid bin. More...
using	bst_idx_t = std::uint64_t
	Type for data row index (sample). More...
using	bst_node_t = std::int32_t
	Type for tree node index. More...
using	bst_group_t = std::uint32_t
	Type for ranking group index. More...
using	bst_target_t = std::uint32_t
	Type for indexing into output targets. More...
using	bst_layer_t = std::int32_t
	Type for indexing boosted layers. More...
using	bst_tree_t = std::int32_t
	Type for indexing trees. More...
using	bst_d_ordinal_t = std::int16_t
	Ordinal of a CUDA device. More...
using	GradientPair = detail::GradientPairInternal< float >
	gradient statistics pair usually needed in gradient boosting More...
using	GradientPairPrecise = detail::GradientPairInternal< double >
	High precision gradient statistics pair. More...
using	Args = std::vector< std::pair< std::string, std::string > >
using	omp_ulong = dmlc::omp_ulong
	define unsigned long for openmp loop More...
using	bst_omp_uint = dmlc::omp_uint
	define unsigned int for openmp loop More...
using	XGBoostVersionT = std::int32_t
	Type used for representing version number in binary form. More...
using	GlobalConfigThreadLocalStore = dmlc::ThreadLocalStore< GlobalConfiguration >
using	F32Array = JsonTypedArray< float, Value::ValueKind::kF32Array >
	Typed UBJSON array for 32-bit floating point. More...
using	F64Array = JsonTypedArray< double, Value::ValueKind::kF64Array >
	Typed UBJSON array for 64-bit floating point. More...
using	I8Array = JsonTypedArray< std::int8_t, Value::ValueKind::kI8Array >
	Typed UBJSON array for int8_t. More...
using	U8Array = JsonTypedArray< std::uint8_t, Value::ValueKind::kU8Array >
	Typed UBJSON array for uint8_t. More...
using	I16Array = JsonTypedArray< std::int16_t, Value::ValueKind::kI16Array >
	Typed UBJSON array for int16_t. More...
using	I32Array = JsonTypedArray< std::int32_t, Value::ValueKind::kI32Array >
	Typed UBJSON array for int32_t. More...
using	I64Array = JsonTypedArray< std::int64_t, Value::ValueKind::kI64Array >
	Typed UBJSON array for int64_t. More...
using	Object = JsonObject
using	Array = JsonArray
using	Number = JsonNumber
using	Integer = JsonInteger
using	Boolean = JsonBoolean
using	String = JsonString
using	Null = JsonNull

Enumerations
enum class	DataType : uint8_t {   kFloat32 = 1 , kDouble = 2 , kUInt32 = 3 , kUInt64 = 4 ,   kStr = 5 }
	data type accepted by xgboost interface More...
enum class	FeatureType : uint8_t { kNumerical = 0 , kCategorical = 1 }
enum class	DataSplitMode : int { kRow = 0 , kCol = 1 }
enum	GPUAccess { kNone , kRead , kWrite }
	Controls data access from the GPU. More...
enum class	PredictionType : std::uint8_t {   kValue = 0 , kMargin = 1 , kContribution = 2 , kApproxContribution = 3 ,   kInteraction = 4 , kApproxInteraction = 5 , kLeaf = 6 }
enum class	MultiStrategy : std::int32_t { kOneOutputPerTree = 0 , kMultiOutputTree = 1 }
	Strategy for building multi-target models. More...

Functions
std::ostream &	operator<< (std::ostream &os, DeviceOrd ord)
template<typename T >
IntrusivePtrCell &	IntrusivePtrRefCount (T const *ptr) noexcept
	User defined function for returning embedded reference count. More...
template<class T , class U >
bool	operator== (IntrusivePtr< T > const &x, IntrusivePtr< U > const &y) noexcept
template<class T , class U >
bool	operator!= (IntrusivePtr< T > const &x, IntrusivePtr< U > const &y) noexcept
template<class T , class U >
bool	operator== (IntrusivePtr< T > const &x, U *y) noexcept
template<class T , class U >
bool	operator!= (IntrusivePtr< T > const &x, U *y) noexcept
template<class T , class U >
bool	operator== (T *x, IntrusivePtr< U > const &y) noexcept
template<class T , class U >
bool	operator!= (T *x, IntrusivePtr< U > const &y) noexcept
template<class T >
bool	operator< (IntrusivePtr< T > const &x, IntrusivePtr< T > const &y) noexcept
template<class T >
bool	operator<= (IntrusivePtr< T > const &x, IntrusivePtr< T > const &y) noexcept
template<class T >
bool	operator> (IntrusivePtr< T > const &x, IntrusivePtr< T > const &y) noexcept
template<class T >
bool	operator>= (IntrusivePtr< T > const &x, IntrusivePtr< T > const &y) noexcept
template<class E , class T , class Y >
std::basic_ostream< E, T > &	operator<< (std::basic_ostream< E, T > &os, IntrusivePtr< Y > const &p)
template<typename T >
bool	IsA (Value const *value)
template<typename T , typename U >
T *	Cast (U *value)
template<typename T >
bool	IsA (Json const &j)
	Check whether a Json object has specific type. More...
template<typename T , typename U >
auto	get (U &json) -> decltype(detail::GetImpl(*Cast< T >(&json.GetValue())))&
	Get Json value. More...
template<typename Parameter >
Object	ToJson (Parameter const &param)
	Convert XGBoost parameter to JSON object. More...
template<typename Parameter >
Args	FromJson (Json const &obj, Parameter *param)
	Load a XGBoost parameter from a JSON object. More...
template<typename T >
T	BuiltinBSwap (T v)
template<typename T , std::enable_if_t< sizeof(T)==1 > * = nullptr>
T	ToBigEndian (T v)
std::ostream &	operator<< (std::ostream &os, StringView const v)
bool	operator== (StringView l, StringView r)
bool	operator!= (StringView l, StringView r)
bool	operator< (StringView l, StringView r)
bool	operator< (std::string const &l, StringView r)
bool	operator< (StringView l, std::string const &r)
StringView	MTNotImplemented ()

Variables
constexpr bst_float	kRtEps = 1e-6f
	small eps gap for minimum split decision. More...

Detailed Description

Core data structure for multi-target trees.

Copyright 2023 by XGBoost contributors

Copyright 2019-2024, XGBoost Contributors

Copyright 2021-2023, XGBoost Contributors

Copyright 2021-2024, XGBoost Contributors

Typedef Documentation

Args

using xgboost::Args = typedef std::vector<std::pair<std::string, std::string> >

Array

using xgboost::Array = typedef JsonArray

Boolean

using xgboost::Boolean = typedef JsonBoolean

bst_bin_t

using xgboost::bst_bin_t = typedef std::int32_t

Type for histogram bin index. We sometimes use -1 to indicate invalid bin.

bst_cat_t

using xgboost::bst_cat_t = typedef std::int32_t

Categorical value type.

bst_d_ordinal_t

using xgboost::bst_d_ordinal_t = typedef std::int16_t

Ordinal of a CUDA device.

bst_feature_t

using xgboost::bst_feature_t = typedef std::uint32_t

Type for data column (feature) index.

bst_float

using xgboost::bst_float = typedef float

float type, used for storing statistics

bst_group_t

using xgboost::bst_group_t = typedef std::uint32_t

Type for ranking group index.

bst_idx_t

using xgboost::bst_idx_t = typedef std::uint64_t

Type for data row index (sample).

bst_layer_t

using xgboost::bst_layer_t = typedef std::int32_t

Type for indexing boosted layers.

bst_node_t

using xgboost::bst_node_t = typedef std::int32_t

Type for tree node index.

bst_omp_uint

using xgboost::bst_omp_uint = typedef dmlc::omp_uint

define unsigned int for openmp loop

bst_target_t

using xgboost::bst_target_t = typedef std::uint32_t

Type for indexing into output targets.

bst_tree_t

using xgboost::bst_tree_t = typedef std::int32_t

Type for indexing trees.

bst_uint

using xgboost::bst_uint = typedef std::uint32_t

unsigned integer type used for feature index.

bst_ulong

using xgboost::bst_ulong = typedef std::uint64_t

unsigned long integers

F32Array

using xgboost::F32Array = typedef JsonTypedArray<float, Value::ValueKind::kF32Array>

Typed UBJSON array for 32-bit floating point.

F64Array

using xgboost::F64Array = typedef JsonTypedArray<double, Value::ValueKind::kF64Array>

Typed UBJSON array for 64-bit floating point.

GlobalConfigThreadLocalStore

using xgboost::GlobalConfigThreadLocalStore = typedef dmlc::ThreadLocalStore<GlobalConfiguration>

GradientPair

using xgboost::GradientPair = typedef detail::GradientPairInternal<float>

gradient statistics pair usually needed in gradient boosting

GradientPairPrecise

using xgboost::GradientPairPrecise = typedef detail::GradientPairInternal<double>

High precision gradient statistics pair.

I16Array

using xgboost::I16Array = typedef JsonTypedArray<std::int16_t, Value::ValueKind::kI16Array>

Typed UBJSON array for int16_t.

I32Array

using xgboost::I32Array = typedef JsonTypedArray<std::int32_t, Value::ValueKind::kI32Array>

Typed UBJSON array for int32_t.

I64Array

using xgboost::I64Array = typedef JsonTypedArray<std::int64_t, Value::ValueKind::kI64Array>

Typed UBJSON array for int64_t.

I8Array

using xgboost::I8Array = typedef JsonTypedArray<std::int8_t, Value::ValueKind::kI8Array>

Typed UBJSON array for int8_t.

Integer

using xgboost::Integer = typedef JsonInteger

Null

using xgboost::Null = typedef JsonNull

Number

using xgboost::Number = typedef JsonNumber

Object

using xgboost::Object = typedef JsonObject

omp_ulong

using xgboost::omp_ulong = typedef dmlc::omp_ulong

define unsigned long for openmp loop

String

using xgboost::String = typedef JsonString

U8Array

using xgboost::U8Array = typedef JsonTypedArray<std::uint8_t, Value::ValueKind::kU8Array>

Typed UBJSON array for uint8_t.

XGBoostVersionT

using xgboost::XGBoostVersionT = typedef std::int32_t

Type used for representing version number in binary form.

Enumeration Type Documentation

DataSplitMode

enum xgboost::DataSplitMode : int

strong

Enumerator
kRow
kCol

DataType

enum xgboost::DataType : uint8_t

strong

data type accepted by xgboost interface

Enumerator
kFloat32
kDouble
kUInt32
kUInt64
kStr

FeatureType

enum xgboost::FeatureType : uint8_t

strong

Enumerator
kNumerical
kCategorical

GPUAccess

enum xgboost::GPUAccess

Controls data access from the GPU.

Since a HostDeviceVector can have data on both the host and device, access control needs to be maintained to keep the data consistent.

There are 3 scenarios supported:

• Data is being manipulated on device. GPU has write access, host doesn't have access.
• Data is read-only on both the host and device.
• Data is being manipulated on the host. Host has write access, device doesn't have access.

Enumerator
kNone
kRead
kWrite

MultiStrategy

enum xgboost::MultiStrategy : std::int32_t

strong

Strategy for building multi-target models.

Enumerator
kOneOutputPerTree
kMultiOutputTree

PredictionType

enum xgboost::PredictionType : std::uint8_t

strong

Enumerator
kValue
kMargin
kContribution
kApproxContribution
kInteraction
kApproxInteraction
kLeaf

Function Documentation

BuiltinBSwap()

template<typename T >

T xgboost::BuiltinBSwap

(

T

v

)

Cast()

template<typename T , typename U >

T* xgboost::Cast

(

U *

value

)

FromJson()

template<typename Parameter >

Args xgboost::FromJson	(	Json const &	obj,
		Parameter *	param
	)

Load a XGBoost parameter from a JSON object.

Template Parameters

Parameter

An instantiation of XGBoostParameter

Parameters

obj	JSON object representing the parameter.
param	Output parameter.

Returns

Unknown arguments in the JSON object.

get()

template<typename T , typename U >

auto xgboost::get

(

U &

json

)

-> decltype(detail::GetImpl(*Cast<T>(&json.GetValue())))&

Get Json value.

Template Parameters

T	One of the Json value type.

Parameters

json

Returns

Value contained in Json object of type T.

IntrusivePtrRefCount()

template<typename T >

IntrusivePtrCell& xgboost::IntrusivePtrRefCount ( T const *  ptr )

noexcept

User defined function for returning embedded reference count.

IsA() [1/2]

template<typename T >

bool xgboost::IsA

(

Json const &

j

)

Check whether a Json object has specific type.

Json json {Array{}};
bool is_array = IsA<Array>(json);
CHECK(is_array);

IsA() [2/2]

template<typename T >

bool xgboost::IsA

(

Value const *

value

)

MTNotImplemented()

StringView xgboost::MTNotImplemented ( )

inline

operator!=() [1/4]

template<class T , class U >

bool xgboost::operator!= ( IntrusivePtr< T > const &  x, IntrusivePtr< U > const &  y  )

noexcept

operator!=() [2/4]

template<class T , class U >

bool xgboost::operator!= ( IntrusivePtr< T > const &  x, U *  y  )

noexcept

operator!=() [3/4]

bool xgboost::operator!= ( StringView  l, StringView  r  )

inline

operator!=() [4/4]

template<class T , class U >

bool xgboost::operator!= ( T *  x, IntrusivePtr< U > const &  y  )

noexcept

operator<() [1/4]

template<class T >

bool xgboost::operator< ( IntrusivePtr< T > const &  x, IntrusivePtr< T > const &  y  )

noexcept

operator<() [2/4]

bool xgboost::operator< ( std::string const &  l, StringView  r  )

inline

operator<() [3/4]

bool xgboost::operator< ( StringView  l, std::string const &  r  )

inline

operator<() [4/4]

bool xgboost::operator< ( StringView  l, StringView  r  )

inline

operator<<() [1/3]

template<class E , class T , class Y >

std::basic_ostream<E, T>& xgboost::operator<<	(	std::basic_ostream< E, T > &	os,
		IntrusivePtr< Y > const &	p
	)

operator<<() [2/3]

std::ostream& xgboost::operator<<	(	std::ostream &	os,
		DeviceOrd	ord
	)

operator<<() [3/3]

std::ostream& xgboost::operator<< ( std::ostream &  os, StringView const  v  )

inline

operator<=()

template<class T >

bool xgboost::operator<= ( IntrusivePtr< T > const &  x, IntrusivePtr< T > const &  y  )

noexcept

operator==() [1/4]

template<class T , class U >

bool xgboost::operator== ( IntrusivePtr< T > const &  x, IntrusivePtr< U > const &  y  )

noexcept

operator==() [2/4]

template<class T , class U >

bool xgboost::operator== ( IntrusivePtr< T > const &  x, U *  y  )

noexcept

operator==() [3/4]

bool xgboost::operator== ( StringView  l, StringView  r  )

inline

operator==() [4/4]

template<class T , class U >

bool xgboost::operator== ( T *  x, IntrusivePtr< U > const &  y  )

noexcept

operator>()

template<class T >

bool xgboost::operator> ( IntrusivePtr< T > const &  x, IntrusivePtr< T > const &  y  )

noexcept

operator>=()

template<class T >

bool xgboost::operator>= ( IntrusivePtr< T > const &  x, IntrusivePtr< T > const &  y  )

noexcept

ToBigEndian()

template<typename T , std::enable_if_t< sizeof(T)==1 > * = nullptr>

T xgboost::ToBigEndian ( T  v )

inline

ToJson()

template<typename Parameter >

Object xgboost::ToJson

(

Parameter const &

param

)

Convert XGBoost parameter to JSON object.

Template Parameters

Parameter

An instantiation of XGBoostParameter

Parameters

param

Input parameter

Returns

JSON object representing the input parameter

Variable Documentation

kRtEps

constexpr bst_float xgboost::kRtEps = 1e-6f

constexpr

small eps gap for minimum split decision.