Struct KernelPCA 

pub struct KernelPCA { /* private fields */ }

Kernel PCA for nonlinear dimensionality reduction

Example

use scirs2_transform::kernel::{KernelPCA, KernelType};
use scirs2_core::ndarray::Array2;

let data = Array2::<f64>::zeros((50, 10));
let mut kpca = KernelPCA::new(2, KernelType::RBF { gamma: 0.1 });
let embedding = kpca.fit_transform(&data).expect("should succeed");
assert_eq!(embedding.shape(), &[50, 2]);

Implementations

impl KernelPCA

pub fn new(n_components: usize, kernel: KernelType) -> Self

Create a new KernelPCA instance

Arguments

• n_components - Number of principal components to retain
• kernel - The kernel function to use

pub fn with_centering(self, center: bool) -> Self

Set whether to center the kernel matrix (default: true)

pub fn kernel(&self) -> &KernelType

Get the kernel type

pub fn eigenvalues(&self) -> Option<&Array1<f64>>

Get the eigenvalues

pub fn explained_variance_ratio(&self) -> Option<&Array1<f64>>

Get the explained variance ratio

pub fn alphas(&self) -> Option<&Array2<f64>>

Get the principal axes (eigenvectors scaled by 1/sqrt(eigenvalue))

pub fn fit<S>(&mut self, x: &ArrayBase<S, Ix2>) -> Result<()>

where S: Data, S::Elem: Float + NumCast,

Fit the Kernel PCA model to the input data

Arguments

• x - Input data, shape (n_samples, n_features)

pub fn transform<S>(&self, x: &ArrayBase<S, Ix2>) -> Result<Array2<f64>>

where S: Data, S::Elem: Float + NumCast,

Transform data using the fitted model

Arguments

• x - Input data, shape (n_samples, n_features)

Returns

• Projected data, shape (n_samples, n_components)

pub fn fit_transform<S>(&mut self, x: &ArrayBase<S, Ix2>) -> Result<Array2<f64>>

where S: Data, S::Elem: Float + NumCast,

Fit and transform in one step

pub fn inverse_transform( &self, projected: &Array2<f64>, max_iter: usize, tol: f64, ) -> Result<Array2<f64>>

Pre-image estimation: approximate reconstruction from kernel space to input space

Uses an iterative fixed-point method (Mika et al., 1998; Kwok & Tsang, 2004) to find an approximate pre-image for the kernel PCA projection.

Arguments

• projected - Projected data in kernel PCA space, shape (n_samples, n_components)
• max_iter - Maximum number of iterations for the fixed-point method
• tol - Convergence tolerance

Returns

• Approximate reconstruction in input space, shape (n_samples, n_features)

pub fn auto_select_gamma<S>( x: &ArrayBase<S, Ix2>, n_components: usize, gamma_values: &[f64], preimage_iter: usize, ) -> Result<(f64, f64)>

where S: Data, S::Elem: Float + NumCast,

Automatic kernel parameter selection via grid search

Tries multiple gamma values and selects the one that minimizes the reconstruction error (for RBF/Laplacian kernels).

Arguments

• x - Input data, shape (n_samples, n_features)
• gamma_values - List of gamma values to try
• preimage_iter - Number of pre-image iterations

Returns

• The best gamma value and corresponding reconstruction error

Trait Implementations

impl Clone for KernelPCA

fn clone(&self) -> KernelPCA

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for KernelPCA

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

Formats the value using the given formatter.