Struct Tensor 

pub struct Tensor<D: Scalar, Mapping: M> { /* private fields */ }

Tensor with scalar type D with axes determined by Mapping.

Implementations

impl<D: Scalar, Mapping: M> Tensor<D, Mapping>

pub fn from_buf(data: Vec<Opt<D>>) -> Self

Creates a new tensor from a buffer.

pub fn to_buf(&self) -> Vec<Opt<D>>

Returns the tensor data as a flat vector in the physical layout order defined by Mapping.

pub fn from_fn<F>(f: F) -> Self

where F: FnMut(&Vec<Term>, &IxDyn) -> Opt<D>,

Creates a new tensor from a function.

pub fn zero() -> Self

Creates a zero tensor.

pub fn rand(rng: &mut impl Rng) -> Self

where StandardUniform: Distribution<D>,

Creates a random tensor.

pub fn uninit() -> Self

Creates an uninitialized tensor.

pub fn view_mut<'l>(&'l mut self) -> TensorViewMut<'l, D, Mapping>

Creates a mutable view of the tensor.

pub fn view<'l>(&'l self) -> TensorView<'l, D, Mapping>

Creates an immutable view of the tensor.

pub fn data_mut(&mut self) -> &mut ArrayD<Opt<D>>

Returns a mutable reference to the underlying data array.

pub unsafe fn transmute<Mapping2: M>(self) -> Tensor<D, Mapping2>

Transmutes the tensor to a different mapping.

Safety

The caller must ensure that the underlying data layout is compatible with the new mapping.

pub unsafe fn reshape<Src: M, Dst: M, Mapping2: M>(self) -> Tensor<D, Mapping2>

Reshapes the tensor to a different mapping. change Src into Dst, final mapping is Mapping2.

Safety

TODO(jeongmin.park); document safety.

pub fn map<D2: Scalar, F>(&self, f: F) -> Tensor<D2, Mapping>

where F: Fn(&Opt<D>) -> Opt<D2>,

Maps the tensor with a function.

pub fn zip_with<D2: Scalar, D3: Scalar, F>( &self, other: &Tensor<D2, Mapping>, f: F, ) -> Tensor<D3, Mapping>

where F: Fn(Opt<D>, Opt<D2>) -> Opt<D3>,

Zips two tensors with a function.

pub fn reduce_add<Dst: M>(&self) -> Tensor<D, Dst>

Performs reduction (sum) over axes not present in Dst.

pub fn reduce<Dst: M>( &self, reduce_fn: impl Fn(Opt<D>, Opt<D>) -> Opt<D>, identity: Opt<D>, ) -> Tensor<D, Dst>

Reduces axes present in self but absent in Dst, using a custom reduce function.

pub fn transpose<Dst: M>(&self, allow_broadcast: bool) -> Tensor<D, Dst>

Performs transpose.

pub fn reduce_then_broadcast<Dst: M>(&self) -> Tensor<D, Dst>

Performs reduction followed by broadcasting to match destination axes. This is useful when the destination has broadcast axes that should be preserved.

Examples

let src: Tensor<f32, m![ABCD]> = Tensor::from_vec(...);  // Shape: ABCD
let dst: Tensor<f32, m![ABE]> = src.reduce_then_broadcast();  // Reduces CD, broadcasts to ABE

pub fn reduce_then_broadcast_with<Dst: M>( &self, reduce_fn: impl Fn(Opt<D>, Opt<D>) -> Opt<D>, identity: Opt<D>, ) -> Tensor<D, Dst>

Performs reduction with a custom function, followed by broadcasting to match destination axes.

pub fn write_scatter<Key: M, Dst: M, Idx: M>( &self, dst: &mut Tensor<D, Dst>, index: &Tensor<i32, Idx>, scaled: bool, )

Scatter elements from self into dst at positions given by index tensor.

When scaled=true, indices are byte offsets; when false, logical positions.

pub fn contraction<Union: M, Lhs: M, Rhs: M>( lhs: &Tensor<D, Lhs>, rhs: &Tensor<D, Rhs>, ) -> Self

Performs contraction between two tensors.

Trait Implementations

impl<D: Clone + Scalar, Mapping: Clone + M> Clone for Tensor<D, Mapping>

fn clone(&self) -> Tensor<D, Mapping>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<D: Debug + Scalar, Mapping: Debug + M> Debug for Tensor<D, Mapping>

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

Formats the value using the given formatter.

impl<D: Scalar, Element: M> From<Tensor<D, Element>> for HostTensor<D, Element>

fn from(inner: Tensor<D, Element>) -> Self

Converts to this type from the input type.

impl<D: VeScalar, Mapping: M> From<Tensor<D, Mapping>> for HasTensor<D, Mapping>

fn from(tensor: Tensor<D, Mapping>) -> Self

Converts to this type from the input type.