Expand description
Regularized statistical models with lazy matrix normalization.
Lasso fits Gaussian lasso models using dense or sparse CSC columns.
Fits return original-scale parameters, reusable training preprocessing,
and explicit convergence diagnostics. See Lasso::terminate_on for the
stopping criterion and LassoFit::termination before using a fit.
The primal objective is P(beta_0, beta) = F(beta_0, beta) + R(beta),
with datafit F, complete penalty R, coefficients beta, and intercept
beta_0. For a feasible dual variable theta, the dual objective is
D(theta) and the duality gap is
G(beta_0, beta, theta) = P(beta_0, beta) - D(theta).
No matrix backend is enabled by default. Versioned features select
LazyMatrix’s integrations with faer 0.24 (faer_v0_24), nalgebra 0.34 and
nalgebra-sparse 0.11 (nalgebra_v0_34), ndarray 0.17 (ndarray_v0_17), or
sprs 0.11 (sprs_v0_11). The short names faer, nalgebra, ndarray, and
sprs are aliases for these features. Match your direct matrix dependency
to the selected release line.
LazyMatrix implements each enabled backend release independently, so
feature unification retains support for the selected matrix types.
faer CSC inputs also accept borrowed SparseColMatRef views, including
cross-validation, without first cloning the source matrix.
With ndarray_v0_17 or its ndarray alias, fit directly against an array or
borrowed view:
use ndarray::array;
use shrinkage::Lasso;
let x = array![[0.0], [1.0], [2.0]];
let fit = Lasso::new(0.1).fit(&x.view(), &[1.0, 3.0, 5.0])?;
let predictions = fit.predict(&x)?;With sprs_v0_11 or its sprs alias, wrap CSC storage in
lazymatrix::SprsCsc. Borrowing a view checks its orientation without copying
its entries:
use shrinkage::{Lasso, lazymatrix::SprsCsc};
use sprs::CsMat;
let x = CsMat::new_csc((3, 1), vec![0, 2], vec![1, 2], vec![1.0, 2.0]);
let columns = SprsCsc::try_new(x.view()).unwrap();
let fit = Lasso::new(0.1).fit(&columns, &[1.0, 3.0, 5.0])?;
let predictions = fit.predict(&columns)?;For CSR input, explicitly convert with to_csc() before wrapping. ndarray
supports row-major, column-major, and strided arrays; column-major storage
keeps each column contiguous for coordinate descent.
lazymatrix::IndexedRows borrows a raw row selection. Normalization uses
only those observations, and predictions retain the original feature scale:
use ndarray::array;
use shrinkage::{Lasso, lazymatrix::IndexedRows};
let x = array![[0.0], [100.0], [2.0]];
let rows = [0, 2];
let training = IndexedRows::new(&x, &rows);
let fit = Lasso::new(0.1).fit(&training, &[1.0, 5.0])?;
assert_eq!(fit.preprocessing().centers(), Some([1.0].as_slice()));
let predictions = fit.predict(&x)?;Indexed columns filter the source on each traversal. Built-in cross-validation instead copies selected entries into owned fold matrices once, keeping repeated solver sweeps on that storage.
Re-exports§
pub use cross_validation::CrossValidation;pub use cross_validation::CrossValidationDatafit;pub use cross_validation::CrossValidationError;pub use cross_validation::CrossValidationFit;pub use cross_validation::FoldAssignments;pub use cross_validation::FoldPath;pub use cross_validation::RowSubset;pub use cross_validation::SelectionRule;pub use cross_validation::ValidationFit;pub use cross_validation::ValidationMetric;pub use datafit::Binomial;pub use datafit::Datafit;pub use datafit::DiagonalHessianDatafit;pub use datafit::DifferentiableDatafit;pub use datafit::Gaussian;pub use datafit::LinkedDatafit;pub use datafit::MultivariateGaussian;pub use datafit::QuadraticModelDatafit;pub use datafit::SmoothDatafit;pub use error::FitError as LassoError;pub use error::BackendError;pub use error::FitError;pub use fit::HybridFit;pub use fit::IrlsFit;pub use fit::LassoFit;pub use fit::MultivariateFit;pub use fit::Preprocessing;pub use fit::ProximalFit;pub use fit::Termination;pub use layout::CoefficientLayout;pub use link::ConvexLink;pub use link::Identity;pub use link::Link;pub use link::Logit;pub use link::Probit;pub use link::TwiceDifferentiableLink;pub use model::Lasso;pub use model::lasso;pub use normalization::Normalization;pub use path::ErasedPathPenalty;pub use path::NullModelDatafit;pub use path::PathDatafit;pub use path::PathFit;pub use path::PathOptions;pub use path::PathPenalty;pub use path::PathSolver;pub use path::PathTermination;pub use path::PenaltyMaximum;pub use path::RegularizationPathSolver;pub use penalty::AxisPenalty;pub use penalty::ChainFusedLasso;pub use penalty::ClusteredLasso;pub use penalty::CoordinatePenalty;pub use penalty::DisjointGroups;pub use penalty::ElasticNet;pub use penalty::GroupLasso;pub use penalty::L1;pub use penalty::Masked;pub use penalty::Penalty;pub use penalty::PenaltyAxis;pub use penalty::ProximalPenalty;pub use penalty::RestrictedProximalPenalty;pub use penalty::Ridge;pub use penalty::Scaled;pub use penalty::Slope;pub use penalty::UncoveredPolicy;pub use problem::MatrixDesign;pub use problem::Problem;pub use problem::RuntimeDesign;pub use problem::RuntimeMultivariateProblem;pub use problem::RuntimePathProblem;pub use problem::RuntimeProblem;pub use screening::DualCertificate;pub use screening::GapSafeDatafit;pub use screening::GapSafePenalty;pub use screening::ScreeningDesign;pub use screening::ScreeningDiagnostics;pub use solver::CoordinateDescent;pub use solver::HybridCoordinateDescent;pub use solver::HybridStep;pub use solver::InitialPoint;pub use solver::Irls;pub use solver::ObservedSolver;pub use solver::ProximalDesign;pub use solver::ProximalGradient;pub use solver::ProximalVector;pub use solver::Solver;pub use solver::SolverObserver;pub use solver::SolverStep;pub use solver::StepKind;pub use solver::StoppingCriterion;pub use solver::WorkingSetDiagnostics;pub use solver::WorkingSetPolicy;pub use lazymatrix;
Modules§
- cross_
validation - Training-only regularization paths, held-out scoring, and model selection.
- datafit
- Loss capabilities on the linear predictor.
- error
- Input, backend, and numerical errors shared by fitting components.
- fit
- Owned fitted parameters, training transformations, and diagnostics.
- layout
- Explicit feature and response axes for CPU coefficient matrices.
- link
- Static link functions and optional second derivatives for GLM composition.
- model
- Named model configurations built from datafits, penalties, and solvers.
- normalization
- Column normalization presets and independent centering and scaling choices.
- path
- Regularization sequences and reusable path-fitting capabilities.
- penalty
- Penalty values, complete proximal maps, and coefficient subproblems.
- problem
- Composition and validated preparation of training problems.
- screening
- Feasible dual certificates and dynamic, fixed-zero restrictions.
- solver
- Solver capabilities, configuration, and stopping rules.
Structs§
- Quantile
Range - Validated probabilities defining a column’s or row’s quantile range.