cvx_analytics::wasserstein

Function sliced_wasserstein

Source 
pub fn sliced_wasserstein(
    a: &[f64],
    b: &[f64],
    centroids: &[&[f32]],
    n_projections: usize,
    seed: u64,
) -> f64
Expand description

Sliced Wasserstein distance between two distributions in K dimensions.

Approximates the true Wasserstein distance by projecting both distributions onto random 1D lines and computing the exact W₁ on each projection. The average over projections converges to the Sliced Wasserstein distance.

§Arguments

  • a - First distribution: weights over K bins (must sum to ~1.0).
  • b - Second distribution: same K bins.
  • centroids - K centroid vectors (one per bin). Used for projection.
  • n_projections - Number of random 1D projections (more = more accurate).
  • seed - Random seed for reproducibility.

§Complexity

O(n_proj × K × (D + K log K)) where D = centroid dimensionality.