xgboost::common::ParallelGroupBuilder< ValueType, SizeType, is_row_major > Class Template Reference

multi-thread version of group builder More...

#include <group_data.h>

Public Member Functions
	ParallelGroupBuilder (std::vector< SizeType > *p_rptr, std::vector< ValueType > *p_data, size_t base_row_offset=0)
	parallel group builder of data.
void	InitBudget (std::size_t max_key, int nthread)
	step 1: initialize the helper, with hint of number keys and thread used in the construction
void	AddBudget (std::size_t key, int threadid, SizeType nelem=1)
	step 2: add budget to each key
void	InitStorage ()
	step 3: initialize the necessary storage
void	Push (std::size_t key, ValueType &&value, int threadid)
	step 4: add data to the allocated space, the calls to this function should be exactly match previous call to AddBudget

Detailed Description

template<typename ValueType, typename SizeType = bst_ulong, bool is_row_major = false>
class xgboost::common::ParallelGroupBuilder< ValueType, SizeType, is_row_major >

multi-thread version of group builder

Template Parameters

ValueType	type of entries in the sparse matrix
SizeType	type of the index range holder
is_row_major	bool value helps to reduce memory for row major

Constructor & Destructor Documentation

ParallelGroupBuilder()

template<typename ValueType , typename SizeType = bst_ulong, bool is_row_major = false>

xgboost::common::ParallelGroupBuilder< ValueType, SizeType, is_row_major >::ParallelGroupBuilder ( std::vector< SizeType > *  p_rptr, std::vector< ValueType > *  p_data, size_t  base_row_offset = 0  )

inline

parallel group builder of data.

Parameters

[in,out]	p_rptr	Row offsets for CSR matrix.
[in,out]	p_data	Data vector to populate
	base_row_offset	(Optional) If the matrix we are building is already partially populated, use this to indicate the row index we are starting from. This saves considerable amounts of time/memory when incrementaly building.

Member Function Documentation

AddBudget()

template<typename ValueType , typename SizeType = bst_ulong, bool is_row_major = false>

void xgboost::common::ParallelGroupBuilder< ValueType, SizeType, is_row_major >::AddBudget ( std::size_t  key, int  threadid, SizeType  nelem = 1  )

inline

step 2: add budget to each key

Parameters

key	the key
threadid	the id of thread that calls this function
nelem	number of element budget add to this row

InitBudget()

template<typename ValueType , typename SizeType = bst_ulong, bool is_row_major = false>

void xgboost::common::ParallelGroupBuilder< ValueType, SizeType, is_row_major >::InitBudget ( std::size_t  max_key, int  nthread  )

inline

step 1: initialize the helper, with hint of number keys and thread used in the construction

Parameters

max_key	number of keys in the matrix, can be smaller than expected, for row major adapter max_key is equal to batch size
nthread	number of thread that will be used in construction

Push()

template<typename ValueType , typename SizeType = bst_ulong, bool is_row_major = false>

void xgboost::common::ParallelGroupBuilder< ValueType, SizeType, is_row_major >::Push ( std::size_t  key, ValueType &&  value, int  threadid  )

inline

step 4: add data to the allocated space, the calls to this function should be exactly match previous call to AddBudget

Parameters

key	the key of group.
value	The value to be pushed to the group.
threadid	the id of thread that calls this function

---

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

• External/xgboost/src/common/group_data.h