Function kmeans

nodedb-vector/src/multivec/plaid.rs:165–186

Run Lloyd's K-means with k-means++ initialisation. Empty clusters are re-seeded each iteration so the result always has exactly `k` distinct centroids covering the input space.

(
    vectors: &[Vec<f32>],
    num_centroids: usize,
    iters: usize,
    seed: u64,
    dim: usize,
)

/// re-seeded each iteration so the result always has exactly `k` distinct
/// centroids covering the input space.
fn kmeans(
    vectors: &[Vec<f32>],
    num_centroids: usize,
    iters: usize,
    seed: u64,
    dim: usize,
) -> Vec<Vec<f32>> {
    if vectors.is_empty() || num_centroids == 0 {
        return Vec::new();
    }

    let k = num_centroids.min(vectors.len());
    let mut centroids = kmeans_plus_plus_init(vectors, k, seed);

    for _ in 0..iters {
        let assignments = assign(vectors, &centroids);
        let new_centroids = recompute(vectors, &assignments, k, dim, &centroids);
        centroids = new_centroids;
    }

    centroids
}

// ---------------------------------------------------------------------------
// PlaidPruner