A type erased view over array_interface protocol defined by numpy. More...

#include <array_interface.h>

Public Member Functions
	ArrayInterface (Object::Map const &array)

	ArrayInterface (Json const &array)

	ArrayInterface (std::string const &str)

	ArrayInterface (StringView str)

void	AssignType (StringView typestr)

XGBOOST_DEVICE std::size_t	Shape (size_t i) const

XGBOOST_DEVICE std::size_t	Stride (size_t i) const

template<typename Fn >
XGBOOST_HOST_DEV_INLINE decltype(auto)	DispatchCall (Fn func) const

XGBOOST_DEVICE std::size_t	ElementSize () const

XGBOOST_DEVICE std::size_t	ElementAlignment () const

template<typename T = float, typename... Index>
XGBOOST_HOST_DEV_INLINE T	operator() (Index &&...index) const

Data Fields
RBitField8	valid

std::size_t	strides [D] {0}

std::size_t	shape [D] {0}

void const *	data {nullptr}

std::size_t	n {0}

bool	is_contiguous {false}

ArrayInterfaceHandler::Type	type {ArrayInterfaceHandler::kF16}

Detailed Description

template<int32_t D, bool allow_mask = (D == 1)>
class xgboost::ArrayInterface< D, allow_mask >

A type erased view over array_interface protocol defined by numpy.

numpy.

Template Parameters

D	The number of maximum dimension.

User input array must have dim <= D for all non-trivial dimensions. During construction, the ctor can automatically remove those trivial dimensions.

Template Parameters

allow_mask Whether masked array is accepted.

Currently this only supported for 1-dim vector, which is used by cuDF column (apache arrow format). For general masked array, as the time of writting, only numpy has the proper support even though it's in the cuda_array_interface protocol defined by numba.

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

External/xgboost/src/data/array_interface.h

Public Member Functions

Data Fields

Detailed Description