Struct ElasticNet 

pub struct ElasticNet<F> {
    pub alpha: F,
    pub l1_ratio: F,
    pub max_iter: usize,
    pub tol: F,
    pub fit_intercept: bool,
    pub positive: bool,
    pub precompute: bool,
    pub selection: CoordSelection,
    pub random_state: Option<u64>,
    pub warm_start: bool,
    pub coef_init: Option<Array1<F>>,
}

ElasticNet regression (L1 + L2 regularized least squares).

Minimizes a combination of L1 and L2 penalties controlled by alpha and l1_ratio. Uses coordinate descent with soft-thresholding to handle the non-smooth L1 component.

Type Parameters

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

Fields

alpha: F

Overall regularization strength. Larger values enforce stronger regularization.

l1_ratio: F

Mix between L1 and L2 regularization.

• l1_ratio = 1.0 → pure Lasso (L1 only)
• l1_ratio = 0.0 → pure Ridge (L2 only)
• 0.0 < l1_ratio < 1.0 → ElasticNet blend

max_iter: usize

Maximum number of coordinate descent iterations.

tol: F

Convergence tolerance on the maximum coefficient change per pass.

fit_intercept: bool

Whether to fit an intercept (bias) term.

positive: bool

When true, constrain coefficients to be non-negative.

precompute: bool

When true, run coordinate descent on the precomputed Gram matrix Q = Xcᵀ Xc and q = Xcᵀ yc instead of recomputing residuals each pass.

Mirrors sklearn ElasticNet(precompute=False) (_coordinate_descent.py:774); the Gram path runs sklearn’s enet_coordinate_descent_gram (_cd_fast.pyx) with the ElasticNet L2 term α·(1−l1_ratio) in the denominator. Reaches the same unique optimum (differing only at floating-point reassociation level, ~1e-10).

selection: CoordSelection

Order in which coordinates are visited each coordinate-descent sweep.

Mirrors sklearn ElasticNet(selection=...) (default Cyclic).

random_state: Option<u64>

Seed for the RNG used when selection == CoordSelection::Random.

Mirrors sklearn ElasticNet(random_state=...) (default None). None falls back to seed 0.

warm_start: bool

When true, initialize coordinate descent from ElasticNet::coef_init (the prior solution) instead of zeros.

Mirrors sklearn ElasticNet(warm_start=False) (_coordinate_descent.py:795), which “reuse[s] the solution of the previous call to fit as initialization” (:796). In sklearn the prior solution is the mutable estimator’s own self.coef_, reused when warm_start is set (_coordinate_descent.py:1062-1063: if not self.warm_start or not hasattr(self, "coef_"): coef_ = np.zeros(...)).

R-DEV-4 adaptation: ferrolearn estimators are immutable value types — there is no mutable self.coef_ carried across repeated .fit() calls. So the prior coefficient vector is supplied EXPLICITLY through ElasticNet::coef_init rather than read off the estimator. The numerics are identical: CD starts from coef_init instead of zeros.

coef_init: Option<Array1<F>>

Explicit coordinate-descent initialization vector used when ElasticNet::warm_start is true (the R-DEV-4 stand-in for sklearn’s reused self.coef_).

Mirrors the coef_init seed fed to the path solver (_coordinate_descent.py:648-651: coef_ = np.zeros(...) when coef_init is None, else coef_ = np.asfortranarray(coef_init, ...)). None (the default) — or warm_start == false — initializes w to zeros, the byte-identical cold-start path. When Some, its length must equal n_features.

Implementations

impl<F: Float + FromPrimitive> ElasticNet<F>

pub fn new() -> Self

Create a new ElasticNet with default settings.

Defaults: alpha = 1.0, l1_ratio = 0.5, max_iter = 1000, tol = 1e-4, fit_intercept = true.

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

Set the overall regularization strength.

pub fn with_l1_ratio(self, l1_ratio: F) -> Self

Set the L1/L2 mixing ratio.

Must be in [0.0, 1.0]. Values outside this range will be rejected at fit time.

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

Set the maximum number of coordinate descent iterations.

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

Set the convergence tolerance on maximum coefficient change.

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

Set whether to fit an intercept term.

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

Set whether to constrain coefficients to be non-negative.

Mirrors sklearn.linear_model.ElasticNet(positive=...).

pub fn with_precompute(self, precompute: bool) -> Self

Set whether to run coordinate descent on the precomputed Gram matrix.

Mirrors sklearn.linear_model.ElasticNet(precompute=...) (_coordinate_descent.py:774); true selects sklearn’s enet_coordinate_descent_gram (_cd_fast.pyx), with the ElasticNet L2 term α·(1−l1_ratio) in the per-coordinate denominator.

pub fn with_selection(self, selection: CoordSelection) -> Self

Set the coordinate-selection order for coordinate descent.

Mirrors sklearn.linear_model.ElasticNet(selection=...).

pub fn with_random_state(self, seed: u64) -> Self

Set the RNG seed used when selection == CoordSelection::Random.

Mirrors sklearn.linear_model.ElasticNet(random_state=...).

pub fn with_warm_start(self, warm_start: bool) -> Self

Enable/disable warm-start coordinate-descent initialization.

Mirrors sklearn.linear_model.ElasticNet(warm_start=...) (_coordinate_descent.py:795): when true, “reuse the solution of the previous call to fit as initialization”. R-DEV-4: ferrolearn estimators are immutable value types with no mutable self.coef_ carried across .fit() calls, so the prior solution is supplied explicitly via ElasticNet::with_coef_init; warm_start only gates whether that vector (when present) is used instead of zeros.

pub fn with_coef_init(self, coef: Array1<F>) -> Self

Provide the explicit coordinate-descent initialization vector used when ElasticNet::warm_start is true.

R-DEV-4 adaptation of sklearn’s reused self.coef_ (_coordinate_descent.py:1062-1063, seeded into the path solver’s coef_init at :648-651): because ferrolearn estimators are immutable value types, the prior coefficient vector is passed in explicitly rather than read off a mutated estimator. Its length must equal n_features at fit time, else Fit::fit returns FerroError::ShapeMismatch.

Trait Implementations

impl<F: Clone> Clone for ElasticNet<F>

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

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<F: Debug> Debug for ElasticNet<F>

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

Formats the value using the given formatter.

impl<F: Float + FromPrimitive> Default for ElasticNet<F>

fn default() -> Self

Returns the “default value” for a type.

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

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

Fit the ElasticNet model using coordinate descent.

Centers the data if fit_intercept is true, then alternates coordinate updates using the soft-threshold rule with L2 scaling.

Errors

• FerroError::ShapeMismatch if x and y have different numbers of samples.
• FerroError::InvalidParameter if alpha is negative, l1_ratio is outside [0, 1], or tol is non-positive.
• FerroError::InsufficientSamples if n_samples == 0.

type Fitted = FittedElasticNet<F>

The fitted model type returned by fit.

type Error = FerroError

The error type returned by fit.

impl<F> PipelineEstimator<F> for ElasticNet<F>

where F: Float + FromPrimitive + ScalarOperand + Send + Sync + 'static,

fn fit_pipeline( &self, x: &Array2<F>, y: &Array1<F>, ) -> Result<Box<dyn FittedPipelineEstimator<F>>, FerroError>

Fit the model and return it as a boxed pipeline estimator.

Errors

Propagates any FerroError from fit.