Struct SparseCOOTensor 

pub struct SparseCOOTensor { /* private fields */ }

Sparse tensor in COO (Coordinate) format.

COO format stores sparse tensors as a pair of tensors:

• indices: shape [sparse_dim, nnz], dtype I64 — the coordinates of non-zero elements
• values: shape [nnz, *dense_dims] — the values at those coordinates

For a sparse tensor with shape [3, 4, 5] and sparse_dim=2:

• indices has shape [2, nnz] (row and column indices)
• values has shape [nnz, 5] (the remaining dense dimension)

The coalesced flag indicates whether duplicate indices have been merged. A coalesced tensor has unique, sorted indices.

Implementations

impl SparseCOOTensor

pub fn new( indices: DenseI64Tensor, values: DenseTensor, dense_shape: Vec<usize>, coalesced: bool, ) -> Result<Self, SparseTensorError>

Create a new sparse COO tensor.

Arguments

• indices - shape [sparse_dim, nnz], dtype I64
• values - shape [nnz, *dense_dims], any floating-point dtype
• dense_shape - the full shape if this were a dense tensor
• coalesced - whether indices are unique and sorted

Errors

Returns error if indices/values shapes don’t match or indices are out of bounds.

pub fn from_coords( coords: &[Vec<i64>], values: Vec<f64>, dense_shape: Vec<usize>, dtype: DType, device: Device, ) -> Result<Self, SparseTensorError>

Create a sparse COO tensor from coordinate lists.

Arguments

• coords - list of coordinate tuples, each of length sparse_dim
• values - flat values for each coordinate
• dense_shape - the full dense shape
• dtype - dtype for the values
• device - device for the tensor

pub fn id(&self) -> u64

pub fn indices(&self) -> &DenseI64Tensor

pub fn values(&self) -> &DenseTensor

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

pub fn sparse_dim(&self) -> usize

pub fn nnz(&self) -> usize

pub fn is_coalesced(&self) -> bool

pub fn dtype(&self) -> DType

pub fn device(&self) -> Device

pub fn version(&self) -> u64

pub fn to_dense(&self) -> Result<DenseTensor, SparseTensorError>

Convert this sparse tensor to a dense tensor.

This allocates a new dense tensor with zeros and fills in the non-zero values.

Trait Implementations

impl Clone for SparseCOOTensor

fn clone(&self) -> SparseCOOTensor

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SparseCOOTensor

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

Formats the value using the given formatter.

impl PartialEq for SparseCOOTensor

fn eq(&self, other: &SparseCOOTensor) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for SparseCOOTensor