Struct FittedKernelPCA 

pub struct FittedKernelPCA<K: Kernel> { /* private fields */ }

A fitted Kernel-PCA model. Stores everything required to project new data into the learned principal subspace: the kernel, the scaled eigenvectors alpha = v / sqrt(lambda), the raw eigenvalues, the training points, and the centering statistics.

Implementations

impl<K: Kernel> FittedKernelPCA<K>

pub fn n_components(&self) -> usize

Number of components kept at fit time.

pub fn feature_dim(&self) -> usize

Feature dimension expected by Self::transform.

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

Eigenvalues of the centered Gram matrix corresponding to the retained components, sorted in descending order.

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

Scaled eigenvectors alpha_k = v_k / sqrt(lambda_k) as an (n_training, n_components) matrix.

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

Borrowed view of the training data retained for projection.

pub fn centering_stats(&self) -> &KernelCenteringStats

Centering statistics captured at fit time.

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

Fraction of centered-Gram variance explained by each retained component. The i-th entry is eigenvalues[i] / sum_j eigenvalues[j]. Returns a zero-length array if the kept eigenvalues are (numerically) zero.

pub fn transform(&self, points: &[Vec<f64>]) -> KernelPcaResult<Array2<f64>>

Project new points into the learned principal subspace.

The returned matrix has shape (points.len(), n_components) and row i is the embedding of points[i].

Errors

• KernelPcaError::DimensionMismatch if any row of points has a different feature dimension than the training data.
• KernelPcaError::InvalidInput if points is empty.

Trait Implementations

impl<K: Kernel> Debug for FittedKernelPCA<K>

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

Formats the value using the given formatter.