Struct CompositionalPca 

pub struct CompositionalPca {
    pub components: Vec<Vec<f64>>,
    pub explained_variance: Vec<f64>,
    pub explained_variance_ratio: Vec<f64>,
    pub clr_mean: Vec<f64>,
    pub n_parts: usize,
    pub n_components: usize,
}

Principal Component Analysis in the Aitchison simplex.

Implemented by applying the CLR transform to each observation and then performing standard PCA on the CLR-transformed data. This is equivalent to PCA in the Aitchison geometry because CLR is an isometry.

References

• Aitchison, J. (1983). Principal component analysis of compositional data. Biometrika, 70(1), 57–65.

Fields

components: Vec<Vec<f64>>

Principal component loadings in CLR space (n_components × D).

explained_variance: Vec<f64>

Explained variance for each component.

explained_variance_ratio: Vec<f64>

Proportion of total variance explained by each component.

clr_mean: Vec<f64>

Column means of CLR-transformed training data (used for centering).

n_parts: usize

Number of parts D.

n_components: usize

Number of components retained.

Implementations

impl CompositionalPca

pub fn fit(data: &[Vec<f64>], n_components: usize) -> StatsResult<Self>

Fit the model on a set of compositional observations.

data: N × D matrix of compositions (each row is one observation). n_components: number of principal components to retain (capped at D−1).

Errors

Returns an error if data is too small or inconsistent.

Examples

use scirs2_stats::compositional::CompositionalPca;
let data = vec![
    vec![0.5, 0.3, 0.2],
    vec![0.4, 0.4, 0.2],
    vec![0.3, 0.5, 0.2],
    vec![0.6, 0.2, 0.2],
];
let pca = CompositionalPca::fit(&data, 2).unwrap();
assert_eq!(pca.n_components, 2);

pub fn transform(&self, data: &[Vec<f64>]) -> StatsResult<Vec<Vec<f64>>>

Project observations onto the principal components.

Each input row is CLR-transformed, centred, and projected. Returns an N × n_components matrix as Vec<Vec<f64>>.

Errors

Returns an error if any row has non-positive components or wrong length.

Examples

use scirs2_stats::compositional::CompositionalPca;
let data = vec![
    vec![0.5, 0.3, 0.2],
    vec![0.4, 0.4, 0.2],
    vec![0.3, 0.5, 0.2],
    vec![0.6, 0.2, 0.2],
];
let pca = CompositionalPca::fit(&data, 2).unwrap();
let scores = pca.transform(&data).unwrap();
assert_eq!(scores.len(), 4);
assert_eq!(scores[0].len(), 2);

pub fn components(&self) -> &[Vec<f64>]

Return the principal component loadings (n_components × D).

pub fn explained_variance(&self) -> &[f64]

Return the explained variance for each component.

Trait Implementations

impl Clone for CompositionalPca

fn clone(&self) -> CompositionalPca

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CompositionalPca

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

Formats the value using the given formatter.