Struct ExclusiveTensor 

pub struct ExclusiveTensor<E, D>
where
    D: Dimension,
{ /* private fields */ }

An exclusive owner of a Tensor and its data.

Given an ordinary Tensor, it is impossible to reason about the lifetime of the data at Tensor::data_ptr. Copies created by Tensor::shallow_clone share the same underlying tensor object and can cause the data memory to be moved or reallocated at any time (for example, by calling Tensor::resize_]).

To avoid this issue, ExclusiveTensor manages the creation of the tensor such that it has exclusive access to the underlying data. An ExclusiveTensor can never provide &Tensor references to the managed tensor.

The managed tensor always lives on the CPU device.

Implementations

impl<E, D> ExclusiveTensor<E, D>

where E: Element, D: Dimension + IntoTorchShape,

pub fn zeros<Sh: IntoDimension<Dim = D>>(shape: Sh) -> Self

Create a zero-initialized tensor.

pub fn ones<Sh: IntoDimension<Dim = D>>(shape: Sh) -> Self

Create a one-initialized tensor.

impl<E, D: Dimension> ExclusiveTensor<E, D>

pub fn into_tensor(self) -> Tensor

Convert into the inner tensor.

impl<E> ExclusiveTensor<E, IxDyn>

where E: Element,

pub fn try_copy_from(tensor: &Tensor) -> Result<Self, ExclusiveTensorError>

Try to create a dynamic-shape tensor by deep copying from a Tensor.

The tensor dtype must match the element type E.

impl<E, D> ExclusiveTensor<E, D>

where E: Element, D: Dimension,

pub fn as_slice(&self) -> &[E]

View the tensor data as a slice.

pub fn as_slice_mut(&mut self) -> &mut [E]

View the tensor data as a mutable slice.

pub fn array_view(&self) -> ArrayView<'_, E, D>

View as an n-dimensional array.

pub fn array_view_mut(&mut self) -> ArrayViewMut<'_, E, D>

View as a mutable n-dimensional array.

Trait Implementations

impl<E: Debug, D> Debug for ExclusiveTensor<E, D>

where D: Dimension + Debug,

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'a, E, D> From<&'a ExclusiveTensor<E, D>> for ArrayView<'a, E, D>

where E: Element, D: Dimension,

fn from(exclusive: &'a ExclusiveTensor<E, D>) -> Self

Converts to this type from the input type.

impl<E, D: Dimension> From<ExclusiveTensor<E, D>> for Tensor

fn from(exclusive: ExclusiveTensor<E, D>) -> Self

Converts to this type from the input type.

impl<A: Element> NumArray1D for ExclusiveTensor<A, Ix1>

type Elem = A

fn zeros(size: usize) -> Self

Create a new instance of the given size with all elements initialized to zero.

fn ones(size: usize) -> Self

Create a new instance of the given size with all elements initialized to one.

fn as_slice(&self) -> &[Self::Elem]

View as a slice.

fn as_slice_mut(&mut self) -> &mut [Self::Elem]

View as a mutable slice.

impl<A: Element> NumArray2D for ExclusiveTensor<A, Ix2>

type Elem = A

fn zeros(size: (usize, usize)) -> Self

Create a new instance of the given size with all elements initialized to zero.

fn ones(size: (usize, usize)) -> Self

Create a new instance of the given size with all elements initialized to one.

fn view(&self) -> ArrayView<'_, Self::Elem, Ix2>

View as a two-dimensional ArrayView.

fn view_mut(&mut self) -> ArrayViewMut<'_, Self::Elem, Ix2>

Mutable view as a two-dimensional ArrayViewMut.