Struct KernelRidgeRegression 

pub struct KernelRidgeRegression { /* private fields */ }

Kernel Ridge Regression

Example

use scirs2_transform::kernel::{KernelRidgeRegression, KernelType};
use scirs2_core::ndarray::{Array1, Array2};

let x = Array2::<f64>::zeros((50, 3));
let y = Array1::<f64>::zeros(50);
let mut krr = KernelRidgeRegression::new(1.0, KernelType::RBF { gamma: 0.1 });
krr.fit(&x, &y).expect("should succeed");
let predictions = krr.predict(&x).expect("should succeed");

Implementations

impl KernelRidgeRegression

pub fn new(alpha: f64, kernel: KernelType) -> Self

Create a new KernelRidgeRegression

Arguments

• alpha - Regularization parameter (lambda). Larger values = more regularization.
• kernel - The kernel function to use

pub fn with_alpha(self, alpha: f64) -> Self

Set the regularization parameter

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

Get the dual coefficients

pub fn kernel(&self) -> &KernelType

Get the kernel type

pub fn regularization(&self) -> f64

Get the regularization parameter

pub fn fit<S1, S2>( &mut self, x: &ArrayBase<S1, Ix2>, y: &ArrayBase<S2, Ix1>, ) -> Result<()>

where S1: Data, S2: Data, S1::Elem: Float + NumCast, S2::Elem: Float + NumCast,

Fit the model with a single output target

Arguments

• x - Training data, shape (n_samples, n_features)
• y - Target values, shape (n_samples,)

pub fn fit_multi<S1, S2>( &mut self, x: &ArrayBase<S1, Ix2>, y: &ArrayBase<S2, Ix2>, ) -> Result<()>

where S1: Data, S2: Data, S1::Elem: Float + NumCast, S2::Elem: Float + NumCast,

Fit the model with multiple output targets

Arguments

• x - Training data, shape (n_samples, n_features)
• y - Target values, shape (n_samples, n_outputs)

pub fn predict<S>(&self, x: &ArrayBase<S, Ix2>) -> Result<Array1<f64>>

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

Predict for new data (single output)

Arguments

• x - Test data, shape (n_test, n_features)

Returns

• Predictions, shape (n_test,) for single output

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

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

Predict for new data (multiple outputs)

Arguments

• x - Test data, shape (n_test, n_features)

Returns

• Predictions, shape (n_test, n_outputs)

pub fn loo_cv(&self) -> Result<(Array2<f64>, f64)>

Leave-one-out cross-validation in closed form

Computes the LOO-CV predictions and error without explicitly re-fitting the model n times. Uses the formula:

LOO_residual_i = alpha_i / (K + lambda*I)^{-1}_{ii}

which requires only one matrix inversion.

Returns

• (loo_predictions, loo_mse) - LOO predictions for each sample and mean squared error

pub fn auto_select_alpha<S1, S2>( x: &ArrayBase<S1, Ix2>, y: &ArrayBase<S2, Ix1>, kernel: &KernelType, alpha_values: &[f64], ) -> Result<(f64, f64)>

where S1: Data, S2: Data, S1::Elem: Float + NumCast, S2::Elem: Float + NumCast,

Automatic selection of the regularization parameter via LOO-CV

Tries multiple alpha values and selects the one with lowest LOO-CV error.

Arguments

• x - Training data
• y - Target values (single output)
• alpha_values - Candidate regularization parameters

Returns

• (best_alpha, best_mse) - Best alpha and corresponding LOO-CV MSE

pub fn score<S>( &self, x: &ArrayBase<S, Ix2>, y_true: &Array1<f64>, ) -> Result<f64>

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

Compute the R-squared score for the training data

Arguments

• y_true - True target values

Returns

• R-squared score

Trait Implementations

impl Clone for KernelRidgeRegression

fn clone(&self) -> KernelRidgeRegression

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for KernelRidgeRegression

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

Formats the value using the given formatter.