Function kmedoids::pammedsil_swap
source · pub fn pammedsil_swap<M, N, L>(
mat: &M,
med: &mut Vec<usize>,
maxiter: usize
) -> (L, Vec<usize>, usize, usize)where
N: Zero + PartialOrd + Copy,
L: Float + Signed + AddAssign + From<N> + From<u32>,
M: ArrayAdapter<N>,
Expand description
Run the original PAMMEDSIL SWAP algorithm (no initialization, but given initial medoids).
This is provided for academic reasons to see the performance difference. Quality-wise, FasterMSC is not worse on average, but much faster. FastMSC is supposed to do the same swaps, and find the same result, but faster.
- type
M
- matrix data type such asndarray::Array2
orkmedoids::arrayadapter::LowerTriangle
- type
N
- number data type such asu32
orf64
- type
L
- number data type such asi64
orf64
for the loss (must be signed) mat
- a pairwise distance matrixmed
- the list of medoidsmaxiter
- the maximum number of iterations allowed
returns a tuple containing:
- the final loss
- the final cluster assignment
- the number of iterations needed
- the number of swaps performed
Panics
- panics when the dissimilarity matrix is not square
- panics when k is 0 or larger than N
Example
Given a dissimilarity matrix of size 4 x 4, use:
let data = ndarray::arr2(&[[0,1,2,3],[1,0,4,5],[2,4,0,6],[3,5,6,0]]);
let mut meds = kmedoids::random_initialization(4, 2, &mut rand::thread_rng());
let (loss, assi, n_iter, n_swap): (f64, _, _, _) = kmedoids::pamsil_swap(&data, &mut meds, 100);
println!("Loss is: {}", loss);