# Function kmedoids::pam_swap

``````pub fn pam_swap<M, N, L>(
mat: &M,
med: &mut Vec<usize>,
maxiter: usize
) -> (L, Vec<usize>, usize, usize)where
N: Zero + PartialOrd + Copy,
L: AddAssign + Signed + Zero + PartialOrd + Copy + From<N>,
Expand description

Run the original PAM SWAP algorithm (no BUILD, but given initial medoids).

This is provided for academic reasons to see the performance difference. Quality-wise, FasterPAM is not worse on average, but much faster. FastPAM1 is supposed to do the same swaps, and find the same result, but faster.

• type `M` - matrix data type such as `ndarray::Array2` or `kmedoids::arrayadapter::LowerTriangle`
• type `N` - number data type such as `u32` or `f64`
• type `L` - number data type such as `i64` or `f64` for the loss (must be signed)
• `mat` - a pairwise distance matrix
• `med` - the list of medoids
• `maxiter` - 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::pam_swap(&data, &mut meds, 100);
println!("Loss is: {}", loss);``````