Struct NamedTensor 

pub struct NamedTensor<T>
where
    T: Float,
{ /* private fields */ }

A Tensor<T> paired with one optional dim name per axis.

names.len() always equals inner.ndim(). Each entry is either Some(String) (the axis has a name) or None (anonymous).

Implementations

impl<T> NamedTensor<T>

where T: Float,

pub fn new( inner: Tensor<T>, names: Vec<Option<String>>, ) -> Result<NamedTensor<T>, FerrotorchError>

Wrap a tensor with explicit per-dim names.

Errors if names.len() != inner.ndim() or two names are equal (and not both None).

pub fn refined( inner: Tensor<T>, names: &[&str], ) -> Result<NamedTensor<T>, FerrotorchError>

Wrap with a flat &[&str] of names; convenient for the common fully-named case. Use "" to leave a dim anonymous.

pub fn tensor(&self) -> &Tensor<T>

Borrow the underlying float tensor. Use this at op boundaries where the named annotation isn’t propagated.

pub fn into_tensor(self) -> Tensor<T>

Consume and return the inner tensor (drops names).

pub fn names(&self) -> &[Option<String>]

Per-dim names (length = ndim).

pub fn shape(&self) -> &[usize]

Logical shape (alias for tensor().shape()).

pub fn ndim(&self) -> usize

Number of dimensions.

pub fn numel(&self) -> usize

Total elements.

pub fn dim_index(&self, name: &str) -> Option<usize>

Index of the dim with the given name, if any.

pub fn size_of(&self, name: &str) -> Option<usize>

Size of the dim with the given name, or None if not found.

pub fn rename( &self, mapping: &[(&str, &str)], ) -> Result<NamedTensor<T>, FerrotorchError>

Replace some dim names according to mapping: [(old, new), ...]. Names not in the mapping are kept as-is. None names are unchanged.

pub fn align_to( &self, target_names: &[&str], ) -> Result<NamedTensor<T>, FerrotorchError>

Return a new NamedTensor whose dims are permuted to match target_names. target_names must contain every named dim of self (and only those); anonymous dims are not allowed in target_names since the permutation is name-driven.

pub fn detached(&self) -> NamedTensor<T>

Drop names and return an unnamed NamedTensor (i.e. all None). Useful before passing into ops that don’t preserve names.

Trait Implementations

impl<T> Clone for NamedTensor<T>

where T: Clone + Float,

fn clone(&self) -> NamedTensor<T>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T> Debug for NamedTensor<T>

where T: Debug + Float,

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

Formats the value using the given formatter.

impl<T> Display for NamedTensor<T>

where T: Float,

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

Formats the value using the given formatter.