Skip to main content

Crate rscopulas

Crate rscopulas 

Source
Expand description

Core copula modeling for validated pseudo-observations.

rscopulas is the main Rust surface for fitting, evaluating, and sampling:

  • single-family copulas such as Gaussian, Student t, Clayton, Frank, and Gumbel-Hougaard,
  • low-level pair-copula kernels with h-functions and inverse h-functions, including Khoudraji asymmetric pair copulas,
  • C-vine, D-vine, and R-vine copulas.

The crate assumes your data is already in pseudo-observation form: finite values strictly inside (0, 1). If you start from raw observations, estimate or transform the marginals first, then build a PseudoObs matrix.

§Quick start

use ndarray::array;
use rand::{rngs::StdRng, SeedableRng};
use rscopulas::{CopulaModel, FitOptions, GaussianCopula, PseudoObs};

let data = PseudoObs::new(array![
    [0.12, 0.18],
    [0.21, 0.25],
    [0.27, 0.22],
    [0.35, 0.42],
    [0.48, 0.51],
    [0.56, 0.49],
    [0.68, 0.73],
    [0.82, 0.79],
])?;

let fit = GaussianCopula::fit(&data, &FitOptions::default())?;
println!("AIC: {}", fit.diagnostics.aic);

let log_pdf = fit.model.log_pdf(&data, &Default::default())?;
println!("first log density = {}", log_pdf[0]);

let mut rng = StdRng::seed_from_u64(7);
let sample = fit.model.sample(4, &mut rng, &Default::default())?;
println!("sample = {:?}", sample);

§Fit a vine copula

use ndarray::array;
use rscopulas::{
    PairCopulaFamily, PseudoObs, SelectionCriterion, VineCopula, VineFitOptions,
};

let data = PseudoObs::new(array![
    [0.12, 0.18, 0.21],
    [0.21, 0.25, 0.29],
    [0.27, 0.22, 0.31],
    [0.35, 0.42, 0.39],
    [0.48, 0.51, 0.46],
    [0.56, 0.49, 0.58],
    [0.68, 0.73, 0.69],
    [0.82, 0.79, 0.76],
])?;

let options = VineFitOptions {
    family_set: vec![
        PairCopulaFamily::Independence,
        PairCopulaFamily::Gaussian,
        PairCopulaFamily::Clayton,
        PairCopulaFamily::Frank,
        PairCopulaFamily::Gumbel,
        PairCopulaFamily::Khoudraji,
    ],
    include_rotations: true,
    criterion: SelectionCriterion::Aic,
    truncation_level: Some(1),
    ..VineFitOptions::default()
};

let fit = VineCopula::fit_r_vine(&data, &options)?;
println!("structure = {:?}", fit.model.structure());
println!("order = {:?}", fit.model.order());

§Backend expectations

The crate exposes explicit execution policy controls through ExecPolicy and Device. Today, Auto is conservative and does not promise that every numerically heavy path uses CUDA or Metal. If you need a deterministic backend choice, prefer ExecPolicy::Force(...).

Re-exports§

pub use data::PseudoObs;
pub use domain::ClaytonCopula;
pub use domain::Copula;
pub use domain::CopulaFamily;
pub use domain::CopulaModel;
pub use domain::Device;
pub use domain::EvalOptions;
pub use domain::ExecPolicy;
pub use domain::FitDiagnostics;
pub use domain::FitOptions;
pub use domain::FrankCopula;
pub use domain::GaussianCopula;
pub use domain::GumbelHougaardCopula;
pub use domain::HacFamily;
pub use domain::HacFitMethod;
pub use domain::HacFitOptions;
pub use domain::HacNode;
pub use domain::HacStructureMethod;
pub use domain::HacTree;
pub use domain::HierarchicalArchimedeanCopula;
pub use domain::SampleOptions;
pub use domain::SelectionCriterion;
pub use domain::StudentTCopula;
pub use domain::VineCopula;
pub use domain::VineEdge;
pub use domain::VineFitOptions;
pub use domain::VineStructure;
pub use domain::VineStructureKind;
pub use domain::VineTree;
pub use errors::BackendError;
pub use errors::CopulaError;
pub use errors::FitError;
pub use errors::InputError;
pub use errors::NumericalError;
pub use paircopula::KhoudrajiParams;
pub use paircopula::PairCopulaFamily;
pub use paircopula::PairCopulaParams;
pub use paircopula::PairCopulaSpec;
pub use paircopula::Rotation;

Modules§

data
domain
errors
fit
math
paircopula
stats
vine