Struct CoClusterResult 

#[non_exhaustive]
pub struct CoClusterResult {
    pub row_labels: Vec<usize>,
    pub col_labels: Vec<usize>,
    pub n_row_blocks: usize,
    pub n_col_blocks: usize,
    pub block_params: Vec<BlockParams>,
    pub row_props: Vec<f64>,
    pub col_props: Vec<f64>,
    pub log_likelihood: f64,
    pub icl: f64,
    pub iterations: usize,
    pub converged: bool,
}

Result of co_cluster.

The block structure is indexed as block_params[k * n_col_blocks + l] where k ∈ 0..n_row_blocks and l ∈ 0..n_col_blocks.

ICL formula

The ICL (Integrated Completed Likelihood) uses the symmetric Govaert-Nadif penalty:

p_KL = (K-1) + (L-1) + 2 * K * L * eff_ncomp
ICL   = log_likelihood - 0.5 * p_KL * (ln(n) + ln(m))

Here ln(n) penalises the n-curve row dimension and ln(m) penalises the m-argument-point column dimension — reflecting that column-clusters partition the m argument points (not the FPC components).

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

row_labels: Vec<usize>

Hard row-cluster assignments, length n. Values in 0..n_row_blocks.

col_labels: Vec<usize>

Hard column-cluster assignments, length m (the number of argument points).

Values in 0..n_col_blocks. This always satisfies col_labels.len() == m — columns cluster the argument points, NOT the FPC components.

n_row_blocks: usize

Number of row clusters K.

n_col_blocks: usize

Number of column clusters L.

block_params: Vec<BlockParams>

Per-block Gaussian parameters, length K*L, indexed k*L + l. Each element describes the diagonal Gaussian on the eff_ncomp-dimensional block scores for the (k, l) block.

row_props: Vec<f64>

Mixing proportions for row clusters, length K. Sums to 1.

col_props: Vec<f64>

Mixing proportions for column clusters, length L. Sums to 1.

log_likelihood: f64

Converged classification log-likelihood (non-decreasing across CEM iterations).

icl: f64

ICL model-selection criterion (finite; lower = better model).

Formula: ICL = log_likelihood - 0.5 * p_KL * (ln(n) + ln(m)) where p_KL = (K-1) + (L-1) + 2*K*L*eff_ncomp.

iterations: usize

Number of CEM iterations performed.

converged: bool

Whether the algorithm converged before max_iter.

Trait Implementations

impl Clone for CoClusterResult

fn clone(&self) -> CoClusterResult

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CoClusterResult

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

Formats the value using the given formatter.

impl PartialEq for CoClusterResult

fn eq(&self, other: &CoClusterResult) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for CoClusterResult