Struct LinearRegression 

pub struct LinearRegression<F> {
    pub fit_intercept: bool,
    pub copy_x: bool,
    pub n_jobs: Option<usize>,
    pub positive: bool,
    /* private fields */
}

Ordinary least squares linear regression.

Solves the least-squares problem via a single SVD (minimum-norm, LAPACK-gelsd-equivalent, through ferray::linalg::lstsq). The fit_intercept option controls whether a bias (intercept) term is included.

Type Parameters

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

Fields

fit_intercept: bool

Whether to fit an intercept (bias) term.

copy_x: bool

Whether X may be overwritten during fit (sklearn copy_X, _base.py:480). ferrolearn’s fit never mutates x (it reads via .iter()/.mean_axis()), so the observable non-mutation contract holds for either value; the field is exposed for ABI parity. Default true, matching sklearn (_base.py:572).

n_jobs: Option<usize>

Number of jobs for the computation (sklearn n_jobs, _base.py:483). ferrolearn’s dense OLS solve is single-threaded, so this is stored but ignored — parallelism is a no-op here and behaviour matches sklearn’s n_jobs=None single-job default. Default None (_base.py:573).

positive: bool

When true, constrains the fitted coefficients to be non-negative via non-negative least squares (sklearn positive, _base.py:574). sklearn solves the (centered, √w-rescaled) coefficient system with scipy.optimize.nnls instead of linalg.lstsq when positive=True (_base.py:645-647). Default false, matching sklearn’s positive=False (_base.py:574); when false, the fit is byte-identical to the unconstrained OLS path.

Implementations

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

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

impl<F: Float> LinearRegression<F>

pub fn new() -> Self

Create a new LinearRegression with default settings.

Defaults: fit_intercept = true, copy_x = true, n_jobs = None, positive = false (mirroring sklearn’s ctor defaults, _base.py:571-574).

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

Set whether to fit an intercept term.

pub fn with_copy_x(self, copy_x: bool) -> Self

Set the copy_X flag (sklearn copy_X, _base.py:480).

ferrolearn’s fit never mutates x, so this is exposed for ABI parity with sklearn and does not change the result.

pub fn with_n_jobs(self, n_jobs: Option<usize>) -> Self

Set the n_jobs parameter (sklearn n_jobs, _base.py:483).

The dense OLS solve is single-threaded; this is stored but ignored.

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

Set the positive flag (sklearn positive, _base.py:574).

When true, the fitted coefficients are constrained to be non-negative, solved via non-negative least squares (scipy.optimize.nnls, _base.py:647) instead of unconstrained OLS. Default false.

Trait Implementations

impl<F: Clone> Clone for LinearRegression<F>

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

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<F: Debug> Debug for LinearRegression<F>

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

Formats the value using the given formatter.

impl<F: Float> Default for LinearRegression<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<F>, Dim<[usize; 1]>>> for LinearRegression<F>

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

Fit the linear regression model.

Solves the OLS least-squares problem via the SVD-based minimum-norm solver [crate::linalg::solve_lstsq] (routed through ferray::linalg::lstsq, LAPACK-gelsd-equivalent), matching scikit-learn’s dense path linalg.lstsq(X, y) (sklearn/linear_model/_base.py:687). When fit_intercept is true, X and y are centered first and the intercept is recovered as y_mean - x_mean . w.

Errors

Returns FerroError::ShapeMismatch if the number of samples in x and y differ. Returns FerroError::InsufficientSamples if there are fewer samples than features. Returns FerroError::NumericalInstability if the system is singular.

type Fitted = FittedLinearRegression<F>

The fitted model type returned by fit.

type Error = FerroError

The error type returned by fit.

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

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

Fit the multi-output linear regression model on a 2-D target Y.

Mirrors scikit-learn’s multi-output dense path: linalg.lstsq(X, Y) with Y of shape (n_samples, n_targets) solves all targets in one SVD, yielding coef_ of shape (n_targets, n_features) and a per-target intercept_ of shape (n_targets,) (sklearn/linear_model/_base.py:687, coef_.T; intercept _set_intercept, _base.py:308/:322). When fit_intercept is true, X and each column of Y are centered by their column means and the intercept is recovered as y_off − coefficients · x_off per target; when false, the solve runs on raw X/Y and the intercepts are all 0.

The 1-D Fit<Array2<F>, Array1<F>> impl is unaffected — this is an additive 2-D arm.

Errors

Returns FerroError::ShapeMismatch if the number of samples in x and y differ. Returns FerroError::InsufficientSamples if there are no samples. Returns FerroError::NumericalInstability if the system is singular.

type Fitted = FittedMultiOutputLinearRegression<F>

The fitted model type returned by fit.

type Error = FerroError

The error type returned by fit.

impl<F> PipelineEstimator<F> for LinearRegression<F>

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

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

Fit this estimator on the given data.