Struct CsfTensor 

pub struct CsfTensor<T> {
    pub shape: Vec<usize>,
    pub mode_order: Vec<usize>,
    pub fib_ptr: Vec<Vec<usize>>,
    pub fib_idx: Vec<Vec<usize>>,
    pub values: Vec<T>,
}

Compressed Sparse Fiber (CSF) tensor.

Stores a sparse tensor of arbitrary order using hierarchical compressed fiber arrays. Modes are ordered as given by mode_order.

Fields

shape: Vec<usize>

Shape of the tensor (one entry per mode).

mode_order: Vec<usize>

Mode ordering: mode_order[level] = which original mode is at this CSF level.

fib_ptr: Vec<Vec<usize>>

Fiber pointers for each non-leaf level. fib_ptr[l] has length = fib_idx[l].len() + 1. Children of fiber i at level l are fib_idx[l+1][fib_ptr[l][i]..fib_ptr[l][i+1]].

fib_idx: Vec<Vec<usize>>

Fiber indices for each level. fib_idx[l] stores the coordinate values at CSF level l.

values: Vec<T>

Values at the leaf level. values[i] corresponds to fib_idx[ndim-1][i].

Implementations

impl<T> CsfTensor<T>

where T: Clone + Copy + Zero + SparseElement + Debug,

pub fn from_coo( indices: &[Vec<usize>], values: &[T], shape: &[usize], mode_order: Option<&[usize]>, ) -> SparseResult<Self>

Construct a CSF tensor from COO-style data (coordinate lists + values).

Arguments

• indices - A slice of index arrays, one per mode. indices[m][i] is the mode-m coordinate of the i-th non-zero.
• values - Non-zero values. Length must match indices[0].len().
• shape - Shape of the tensor.
• mode_order - Permutation of 0..ndim specifying which tensor mode goes at each CSF level. Pass None for natural order.

pub fn ndim(&self) -> usize

Number of dimensions.

pub fn nnz(&self) -> usize

Number of stored non-zeros.

pub fn get(&self, indices: &[usize]) -> Option<T>

Look up a value by original tensor coordinates.

Returns T::sparse_zero() if the entry is not stored.

pub fn fiber( &self, free_mode: usize, fixed_indices: &[usize], ) -> SparseResult<Vec<(usize, T)>>

Extract a fiber from the tensor.

A fiber is a 1D slice obtained by fixing all indices except one mode. Returns (index, value) pairs for the free mode.

Arguments

• free_mode - The mode to leave free (tensor mode index, not CSF level).
• fixed_indices - Values for all other modes. Length = ndim - 1. The order corresponds to modes 0, 1, ..., free_mode-1, free_mode+1, ..., ndim-1.

pub fn matricize( &self, mode: usize, ) -> SparseResult<(Vec<usize>, Vec<usize>, Vec<T>)>

Mode-n matricization: unfold the tensor along mode n.

Returns a matrix (in COO form as (rows, cols, vals)) where:

• rows correspond to the free mode’s indices
• columns correspond to the lexicographic combination of all other mode indices

The column index for a multi-index (i_0, ..., i_{n-1}, i_{n+1}, ..., i_{N-1}) is computed as a mixed-radix number.

pub fn memory_usage(&self) -> usize

Memory usage estimate in bytes.

Trait Implementations

impl<T: Clone> Clone for CsfTensor<T>

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

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T: Debug> Debug for CsfTensor<T>

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

Formats the value using the given formatter.