Struct CCIPCA 

pub struct CCIPCA { /* private fields */ }

Candid Covariance-free Incremental PCA (CCIPCA).

Maintains n_components eigenvector estimates that are updated with each new sample using a covariance-free rank-one update rule. The amnestic weight controls how quickly old data is forgotten (0.0 = equal weighting).

Dimensionality is determined lazily from the first call to update, matching the pattern of IncrementalNormalizer.

Implementations

impl CCIPCA

pub fn new(n_components: usize) -> Self

Create a CCIPCA estimator with n_components components and no forgetting.

use irithyll::preprocessing::CCIPCA;
let pca = CCIPCA::new(3);
assert_eq!(pca.n_components(), 3);
assert_eq!(pca.count(), 0);

pub fn with_amnestic(n_components: usize, amnestic: f64) -> Self

Create a CCIPCA estimator with a custom amnestic (forgetting) weight.

Higher amnestic values give more weight to recent samples. A value of 0.0 weights all samples equally.

use irithyll::preprocessing::CCIPCA;
let pca = CCIPCA::with_amnestic(2, 1.5);
assert_eq!(pca.n_components(), 2);

pub fn update(&mut self, features: &[f64])

Incorporate a new sample and update eigencomponent estimates.

On the first call, the feature dimensionality is established from features.len(). Subsequent calls must provide the same length.

The algorithm performs:

1. Welford mean update.
2. Center the sample.
3. Rank-one update of each eigenvector with re-orthogonalisation.
4. Sort components by descending eigenvalue.

Panics

Panics if features.len() does not match the established dimensionality.

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

Project features onto the current principal components.

Centers the input using the running mean and computes the dot product with each eigenvector. Returns a Vec of length n_components.

Panics

Panics if features.len() does not match the established dimensionality, or if no samples have been incorporated yet.

pub fn update_and_transform(&mut self, features: &[f64]) -> Vec<f64>

Update eigencomponents with features, then project onto them.

Equivalent to calling update followed by transform.

pub fn reset(&mut self)

Reset all state, preserving n_components and amnestic settings.

After reset, the next update call will re-initialise the feature dimensionality from the input.

pub fn n_components(&self) -> usize

Number of principal components being estimated.

pub fn n_features(&self) -> Option<usize>

Feature dimensionality, or None if not yet initialised.

pub fn count(&self) -> u64

Number of samples incorporated so far.

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

Current eigenvalue estimates (sorted descending).

pub fn explained_variance_ratio(&self) -> Vec<f64>

Explained variance ratio for each component.

Each element is the component’s eigenvalue divided by the sum of all eigenvalues. Returns an empty Vec if no samples have been incorporated.

Trait Implementations

impl Clone for CCIPCA

fn clone(&self) -> CCIPCA

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for CCIPCA

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

Formats the value using the given formatter.

impl Default for CCIPCA

fn default() -> Self

Default CCIPCA with 2 components and no forgetting.

impl StreamingPreprocessor for CCIPCA

fn update_and_transform(&mut self, features: &[f64]) -> Vec<f64>

Update internal statistics from this sample and return transformed features.

fn transform(&self, features: &[f64]) -> Vec<f64>

Transform features using current statistics without updating them.

fn output_dim(&self) -> Option<usize>

Number of output features, or None if unknown until the first sample.

fn reset(&mut self)

Reset to initial (untrained) state.