The primary data structure in NEML2 for working with labeled tensor views. More...

Detailed Description

template<class Derived, Size D>
class neml2::LabeledTensor< Derived, D >

The primary data structure in NEML2 for working with labeled tensor views.

Each LabeledTensor consists of one Tensor and one or more LabeledAxis. The LabeledTensor<D> is templated on the base dimension \(D\). LabeledTensor handles the creation, modification, and accessing of labeled tensors.

Template Parameters

D	The number of base dimensions

#include <LabeledTensor.h>

Public Member Functions
	LabeledTensor ()=default
	Default constructor.

	LabeledTensor (const torch::Tensor &tensor, const std::array< const LabeledAxis *, D > &axes)
	Construct from a torch::Tensor and array of `LabeledAxis`

	LabeledTensor (const Tensor &tensor, const std::array< const LabeledAxis *, D > &axes)
	Construct from a Tensor with array of `LabeledAxis`

	LabeledTensor (const Derived &other)
	Copy constructor.

LabeledTensor< Derived, D > &	operator= (const Derived &other)
	Assignment operator.

template<typename T , typename = std::enable_if_t<std::is_base_of_v<TensorBase<T>, T>>>
T	reinterpret (indexing::TensorLabelsRef indices) const
	Get a tensor by slicing on the base dimensions AND reinterpret it as a primitive tensor.

const std::array< const LabeledAxis *, D > &	axes () const
	Get all the labeled axes.

const LabeledAxis &	axis (Size i=0) const
	Get a specific labeled axis.


	operator Tensor () const

	operator torch::Tensor () const


const Tensor &	tensor () const

Tensor &	tensor ()

Meta operations
Derived	clone (torch::MemoryFormat memory_format=torch::MemoryFormat::Contiguous) const

Derived	detach () const
	Return a copy without gradient graphs.

void	detach_ ()
	Detach from gradient graphs.

Derived	to (const torch::TensorOptions &options) const
	Change tensor options.

void	copy_ (const torch::Tensor &other)
	Copy another tensor.

void	zero_ ()
	Set all entries to zero.

bool	requires_grad () const
	Get the requires_grad property.

void	requires_grad_ (bool req=true)
	Set the requires_grad property.

Derived	operator- () const
	Negation.

Tensor information
torch::TensorOptions	options () const

torch::Dtype	scalar_type () const
	Tensor options.

torch::Device	device () const
	Tensor options.

Size	dim () const
	Number of tensor dimensions.

TensorShapeRef	sizes () const
	Tensor shape.

Size	size (Size dim) const
	Tensor shape.

bool	batched () const
	Whether the tensor is batched.

Size	batch_dim () const
	Return the number of batch dimensions.

constexpr Size	base_dim () const
	Return the number of base dimensions.

TensorShapeRef	batch_sizes () const
	Return the batch size.

Size	batch_size (Size d) const
	Return the length of some batch axis.

TensorShapeRef	base_sizes () const
	Return the base size.

Size	base_size (Size d) const
	Return the length of some base axis.

Size	base_storage () const
	Return the flattened storage needed just for the base indices.

Getter and setter
Derived	batch_index (indexing::TensorIndicesRef indices) const

Tensor	base_index (indexing::TensorLabelsRef labels) const
	Get a tensor by slicing on the base dimensions.

void	batch_index_put_ (indexing::TensorIndicesRef indices, const torch::Tensor &other)
	Set values by slicing on the batch dimensions.

void	base_index_put_ (indexing::TensorLabelsRef labels, const Tensor &other)
	Set values by slicing on the base dimensions.

Static Public Member Functions
static Derived	empty (TensorShapeRef batch_shape, const std::array< const LabeledAxis *, D > &axes, const torch::TensorOptions &options=default_tensor_options())
	Setup new empty storage.

static Derived	zeros (TensorShapeRef batch_shape, const std::array< const LabeledAxis *, D > &axes, const torch::TensorOptions &options=default_tensor_options())
	Setup new storage with zeros.

Protected Attributes
Tensor	_tensor
	The tensor.

std::array< const LabeledAxis *, D >	_axes
	The labeled axes of this tensor.

Constructor & Destructor Documentation

◆ LabeledTensor() [1/4]

template<class Derived , Size D>

LabeledTensor ( )

default

Default constructor.

◆ LabeledTensor() [2/4]

template<class Derived , Size D>

LabeledTensor	(	const torch::Tensor &	tensor,
		const std::array< const LabeledAxis *, D > &	axes )

Construct from a torch::Tensor and array of LabeledAxis

◆ LabeledTensor() [3/4]

template<class Derived , Size D>

LabeledTensor	(	const Tensor &	tensor,
		const std::array< const LabeledAxis *, D > &	axes )

Construct from a Tensor with array of LabeledAxis

◆ LabeledTensor() [4/4]

template<class Derived , Size D>

LabeledTensor ( const Derived & other )

Copy constructor.

Member Function Documentation

◆ axes()

template<class Derived , Size D>

const std::array< const LabeledAxis *, D > & axes ( ) const

inline

Get all the labeled axes.

◆ axis()

template<class Derived , Size D>

const LabeledAxis & axis ( Size i = 0 ) const

inline

Get a specific labeled axis.

◆ base_dim()

template<class Derived , Size D>

constexpr Size base_dim ( ) const

inlineconstexpr

Return the number of base dimensions.

◆ base_index()

template<class Derived , Size D>

Tensor base_index ( indexing::TensorLabelsRef labels ) const

Get a tensor by slicing on the base dimensions.

◆ base_index_put_()

template<class Derived , Size D>

void base_index_put_	(	indexing::TensorLabelsRef	labels,
		const Tensor &	other )

Set values by slicing on the base dimensions.

◆ base_size()

template<class Derived , Size D>

Size base_size ( Size d ) const

Return the length of some base axis.

◆ base_sizes()

template<class Derived , Size D>

TensorShapeRef base_sizes ( ) const

Return the base size.

◆ base_storage()

template<class Derived , Size D>

Size base_storage ( ) const

Return the flattened storage needed just for the base indices.

◆ batch_dim()

template<class Derived , Size D>

Size batch_dim ( ) const

Return the number of batch dimensions.

◆ batch_index()

template<class Derived , Size D>

Derived batch_index ( indexing::TensorIndicesRef indices ) const

Get a tensor by slicing on the batch dimensions

◆ batch_index_put_()

template<class Derived , Size D>

void batch_index_put_	(	indexing::TensorIndicesRef	indices,
		const torch::Tensor &	other )

Set values by slicing on the batch dimensions.

◆ batch_size()

template<class Derived , Size D>

Size batch_size ( Size d ) const

Return the length of some batch axis.

◆ batch_sizes()

template<class Derived , Size D>

TensorShapeRef batch_sizes ( ) const

Return the batch size.

◆ batched()

template<class Derived , Size D>

bool batched ( ) const

Whether the tensor is batched.

◆ clone()

template<class Derived , Size D>

Derived clone ( torch::MemoryFormat memory_format = torch::MemoryFormat::Contiguous ) const

Clone this LabeledTensor

◆ copy_()

template<class Derived , Size D>

void copy_ ( const torch::Tensor & other )

Copy another tensor.

◆ detach()

template<class Derived , Size D>

Derived detach ( ) const

Return a copy without gradient graphs.

◆ detach_()

template<class Derived , Size D>

void detach_ ( )

Detach from gradient graphs.

◆ device()

template<class Derived , Size D>

torch::Device device ( ) const

Tensor options.

◆ dim()

template<class Derived , Size D>

Size dim ( ) const

Number of tensor dimensions.

◆ empty()

template<class Derived , Size D>

Derived empty	(	TensorShapeRef	batch_shape,
		const std::array< const LabeledAxis *, D > &	axes,
		const torch::TensorOptions &	options = default_tensor_options() )

static

Setup new empty storage.

◆ operator Tensor()

template<class Derived , Size D>

operator Tensor ( ) const

Implicit conversion

◆ operator torch::Tensor()

template<class Derived , Size D>

operator torch::Tensor ( ) const

◆ operator-()

template<class Derived , Size D>

Derived operator- ( ) const

Negation.

◆ operator=()

template<class Derived , Size D>

LabeledTensor< Derived, D > & operator= ( const Derived & other )

Assignment operator.

◆ options()

template<class Derived , Size D>

torch::TensorOptions options ( ) const

Tensor options

◆ reinterpret()

template<class Derived , Size D>

template<typename T , typename >

T reinterpret ( indexing::TensorLabelsRef indices ) const

Get a tensor by slicing on the base dimensions AND reinterpret it as a primitive tensor.

◆ requires_grad()

template<class Derived , Size D>

bool requires_grad ( ) const

Get the requires_grad property.

◆ requires_grad_()

template<class Derived , Size D>

void requires_grad_ ( bool req = true )

Set the requires_grad property.

◆ scalar_type()

template<class Derived , Size D>

torch::Dtype scalar_type ( ) const

Tensor options.

◆ size()

template<class Derived , Size D>

Size size ( Size dim ) const

Tensor shape.

◆ sizes()

template<class Derived , Size D>

TensorShapeRef sizes ( ) const

Tensor shape.

◆ tensor() [1/2]

template<class Derived , Size D>

Tensor & tensor ( )

inline

◆ tensor() [2/2]

template<class Derived , Size D>

const Tensor & tensor ( ) const

inline

Get the underlying tensor

◆ to()

template<class Derived , Size D>

Derived to ( const torch::TensorOptions & options ) const

Change tensor options.

◆ zero_()

template<class Derived , Size D>

void zero_ ( )

Set all entries to zero.

◆ zeros()

template<class Derived , Size D>

Derived zeros	(	TensorShapeRef	batch_shape,
		const std::array< const LabeledAxis *, D > &	axes,
		const torch::TensorOptions &	options = default_tensor_options() )

static

Setup new storage with zeros.

Member Data Documentation

◆ _axes

template<class Derived , Size D>

std::array<const LabeledAxis *, D> _axes

protected

The labeled axes of this tensor.

◆ _tensor

template<class Derived , Size D>

Tensor _tensor

protected

The tensor.

Detailed Description

Public Member Functions

Static Public Member Functions

Protected Attributes

Constructor & Destructor Documentation

◆ LabeledTensor() [1/4]

◆ LabeledTensor() [2/4]

◆ LabeledTensor() [3/4]

◆ LabeledTensor() [4/4]

Member Function Documentation

◆ axes()

◆ axis()

◆ base_dim()

◆ base_index()

◆ base_index_put_()

◆ base_size()

◆ base_sizes()

◆ base_storage()

◆ batch_dim()

◆ batch_index()

◆ batch_index_put_()

◆ batch_size()

◆ batch_sizes()

◆ batched()

◆ clone()

◆ copy_()

◆ detach()

◆ detach_()

◆ device()

◆ dim()

◆ empty()

◆ operator Tensor()

◆ operator torch::Tensor()

◆ operator-()

◆ operator=()

◆ options()

◆ reinterpret()

◆ requires_grad()

◆ requires_grad_()

◆ scalar_type()

◆ size()

◆ sizes()

◆ tensor() [1/2]

◆ tensor() [2/2]

◆ to()

◆ zero_()

◆ zeros()

Member Data Documentation

◆ _axes

◆ _tensor