Struct RidgeClassifier 

pub struct RidgeClassifier<F> {
    pub alpha: F,
    pub fit_intercept: bool,
    pub max_iter: Option<usize>,
    pub tol: F,
    pub positive: bool,
    pub class_weight: ClassWeight<F>,
}

Ridge Classifier.

Applies Ridge regression (L2-regularized least squares) to classification by converting labels to a binary indicator matrix.

Type Parameters

• F: The floating-point type (f32 or f64).

Fields

alpha: F

Regularization strength. Larger values specify stronger regularization.

fit_intercept: bool

Whether to fit an intercept (bias) term.

max_iter: Option<usize>

Maximum number of iterations for iterative solvers (sklearn max_iter, _ridge.py:1520). Exposed for sklearn ABI parity; ferrolearn implements only the direct dense solver, so this field is stored but has no effect on the computed result. Default None, matching sklearn’s default (_ridge.py:1520).

tol: F

Convergence tolerance for iterative solvers (sklearn tol, _ridge.py:1521). Exposed for sklearn ABI parity; ferrolearn implements only the direct dense solver, so this field is stored but has no effect on the computed result. Default 1e-4, matching sklearn’s default (_ridge.py:1521).

positive: bool

When true, constrain the coefficients to be non-negative (sklearn positive, _ridge.py:902/:911). The per-target indicator-ridge solve is routed through projected coordinate descent (crate::linalg::nonneg_ridge_cd) using max_iter/tol, mirroring the optimum sklearn reaches with its L-BFGS-B solver for positive=True (_ridge.py:329). Default false, matching sklearn (_ridge.py:902).

class_weight: ClassWeight<F>

Per-class sample weighting (sklearn class_weight, _ridge.py:1398). Reweights samples by class membership before the weighted ridge fit, multiplying into any user sample_weight (_ridge.py:1305-1307, mirroring sklearn.utils.class_weight.compute_class_weight). Default ClassWeight::None (all classes weight 1.0, matching sklearn class_weight=None, _ridge.py:1400).

Implementations

impl<F: Float> RidgeClassifier<F>

pub fn new() -> Self

Create a new RidgeClassifier with default settings.

Defaults: alpha = 1.0, fit_intercept = true, max_iter = None, tol = 1e-4, positive = false — mirroring sklearn’s ctor defaults (sklearn/linear_model/_ridge.py:1514-1526, positive=False _ridge.py:902).

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

Set the regularization strength.

pub fn with_fit_intercept(self, fit_intercept: bool) -> Self

Set whether to fit an intercept term.

pub fn with_max_iter(self, max_iter: Option<usize>) -> Self

Set the maximum number of iterations for iterative solvers (sklearn max_iter, _ridge.py:1520).

ferrolearn’s direct solver solves in closed form with no iteration, so this is stored for sklearn ABI parity and does not affect the computed result.

pub fn with_tol(self, tol: F) -> Self

Set the convergence tolerance for iterative solvers (sklearn tol, _ridge.py:1521).

ferrolearn’s direct solver solves in closed form with no iteration, so this is stored for sklearn ABI parity and does not affect the computed result.

pub fn with_positive(self, positive: bool) -> Self

Constrain the fitted coefficients to be non-negative (sklearn positive, _ridge.py:902/:911).

When true, each indicator-target ridge solve is routed through projected coordinate descent (crate::linalg::nonneg_ridge_cd) using max_iter/tol, yielding the same optimum sklearn reaches with its L-BFGS-B solver for positive=True (_ridge.py:329). false (default) is byte-identical to the unconstrained closed-form path.

pub fn with_class_weight(self, class_weight: ClassWeight<F>) -> Self

Set the per-class sample weighting strategy (sklearn class_weight, _ridge.py:1398).

The selected per-class weight reweights samples by class membership and multiplies into any user sample_weight before the weighted ridge fit (_ridge.py:1305-1307, mirroring sklearn.utils.class_weight.compute_class_weight). ClassWeight::None (default) leaves the unweighted fit byte-identical.

impl<F: Float + Send + Sync + ScalarOperand + FromPrimitive + LinalgFloat + 'static> RidgeClassifier<F>

pub fn fit_with_sample_weight( &self, x: &Array2<F>, y: &Array1<usize>, sample_weight: Option<&Array1<F>>, ) -> Result<FittedRidgeClassifier<F>, FerroError>

Fit the Ridge Classifier with optional per-sample weights.

Mirrors scikit-learn’s RidgeClassifier.fit(X, y, sample_weight=None) (sklearn/linear_model/_ridge.py:1220). The target is encoded as an indicator matrix Y (binary {-1, +1} single column, multiclass one-hot) via LabelBinarizer(pos_label=1, neg_label=-1) (_ridge.py:1300-1301), then the underlying weighted ridge is solved on (X, Y): sample_weight is forwarded into _BaseRidge.fit which does weighted _preprocess_data (weighted offsets) + _rescale_data (√w row-rescale, _ridge.py:682-688).

When sample_weight is Some(w) and fit_intercept is true, the weighted offsets x_off[j] = Σᵢ wᵢ·x[i,j] / Σwᵢ, y_off[t] = Σᵢ wᵢ·Y[i,t] / Σwᵢ center X/Y, then each row is rescaled by √wᵢ before the per-target ridge solve, and intercept[t] = y_off[t] − Σⱼ x_off[j]·coef[j,t]. With fit_intercept false only the √w row-rescale is applied and the intercept is 0.

sample_weight = None is BYTE-IDENTICAL to Fit::fit (the unweighted centering + per-target linalg::solve_ridge path).

Errors

• FerroError::ShapeMismatch — sample count mismatch, or sample_weight.len() != n_samples.
• FerroError::InvalidParameter — negative alpha.
• FerroError::InsufficientSamples — fewer than 2 classes / no samples.

Trait Implementations

impl<F: Clone> Clone for RidgeClassifier<F>

fn clone(&self) -> RidgeClassifier<F>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<F: Debug> Debug for RidgeClassifier<F>

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

Formats the value using the given formatter.

impl<F: Float> Default for RidgeClassifier<F>

fn default() -> Self

Returns the “default value” for a type.

impl<F: Float + Send + Sync + ScalarOperand + FromPrimitive + LinalgFloat + 'static> Fit<ArrayBase<OwnedRepr<F>, Dim<[usize; 2]>>, ArrayBase<OwnedRepr<usize>, Dim<[usize; 1]>>> for RidgeClassifier<F>

fn fit( &self, x: &Array2<F>, y: &Array1<usize>, ) -> Result<FittedRidgeClassifier<F>, FerroError>

Fit the Ridge Classifier by converting labels to a binary indicator matrix and solving multivariate Ridge regression.

Equivalent to RidgeClassifier::fit_with_sample_weight with sample_weight = None.

Errors

• FerroError::ShapeMismatch — sample count mismatch.
• FerroError::InvalidParameter — negative alpha.
• FerroError::InsufficientSamples — fewer than 2 classes.

type Fitted = FittedRidgeClassifier<F>

The fitted model type returned by fit.

type Error = FerroError

The error type returned by fit.