Struct DiffSortNet 

pub struct DiffSortNet {
    pub network_type: NetworkType,
    pub size: usize,
    pub steepness: f64,
    pub dist: RelaxDist,
    /* private fields */
}

A differentiable sorting network.

Fields

network_type: NetworkType

Network topology type.

size: usize

Input size (padded to power-of-2 for bitonic).

steepness: f64

Inverse temperature (steepness). Higher = sharper, closer to hard sort.

dist: RelaxDist

Distribution for the relaxed comparator.

Implementations

impl DiffSortNet

pub fn new( network_type: NetworkType, size: usize, steepness: f64, dist: RelaxDist, ) -> Self

Create a new differentiable sorting network.

For Bitonic, the size must be a power of 2. If not, inputs will be padded with -inf to the next power of 2.

Arguments

• network_type - Bitonic or OddEven
• size - Number of elements to sort
• steepness - Inverse temperature (higher = sharper)
• dist - Distribution family for comparator relaxation

pub fn sort(&self, x: &[f64]) -> Result<(Vec<f64>, Vec<Vec<f64>>)>

Sort the input, returning (sorted_values, soft_permutation_matrix).

The permutation matrix P satisfies: sorted ≈ x @ P (row convention). P[i][j] is the soft weight of input position i contributing to output position j.

pub fn num_comparators(&self) -> usize

Return the number of comparators in this sorting network.

pub fn comparator_pairs(&self) -> &[(usize, usize)]

Return the comparator pairs as a slice.

Each pair (a, b) means “min at a, max at b”.

pub fn depth(&self) -> usize

Return the network depth (number of parallel stages).

Trait Implementations

impl Clone for DiffSortNet

fn clone(&self) -> DiffSortNet

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DiffSortNet

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

Formats the value using the given formatter.