| _version_ | 1866902235811676160 |
|---|---|
| author | Agostino Gnasso |
| author_facet | Agostino Gnasso |
| contents | <h2>shapCopula v0.1.0</h2> <p>R implementation accompanying the paper:</p> <blockquote> <p>Gnasso, A. (2026). <em>Inference for Conditional Shapley Values via Vine Copulas</em>. Working paper.</p> </blockquote> <h3>What this package does</h3> <p><code>shapCopula</code> computes <strong>Conditional SAGE</strong> (Shapley Additive Global importancE) values for any black-box predictor, with asymptotically valid Wald confidence intervals and hypothesis tests for global feature importance.</p> <p>The conditional value function E[f(X) | X_S = x_S] is estimated via a <strong>Gaussian copula with empirical marginals</strong> (default) or a <strong>nonparametric vine copula</strong> (<code>rvinecopulib</code>). A one-step debiased estimator with K-fold cross-fitting and U-statistic debiasing of the squared loss yields root-n consistent, asymptotically normal SAGE estimates (Theorem 1 in the paper).</p> <h3>Main functions</h3> <p>| Function | Description | |----------|-------------| | <code>sage_cv_gcop()</code> | Conditional SAGE via Gaussian copula — <strong>recommended</strong> | | <code>sage_cv()</code> | Conditional SAGE via nonparametric vine copula | | <code>sage_cv_marginal()</code> | Marginal (interventional) SAGE — baseline | | <code>fit_gauss_copula()</code> | Gaussian copula fitting with empirical marginals | | <code>fit_copula()</code> | Nonparametric vine copula fitting (<code>rvinecopulib</code>) |</p> <h3>Installation</h3> <pre><code class="language-r">remotes::install_github("agostinognasso/shapCopula") Requires: R ≥ 4.1.0, rvinecopulib ≥ 0.6.0 --- </code></pre> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_20392904 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | agostinognasso/shapCopula: shapCopula v0.1.0 — Initial release Agostino Gnasso <h2>shapCopula v0.1.0</h2> <p>R implementation accompanying the paper:</p> <blockquote> <p>Gnasso, A. (2026). <em>Inference for Conditional Shapley Values via Vine Copulas</em>. Working paper.</p> </blockquote> <h3>What this package does</h3> <p><code>shapCopula</code> computes <strong>Conditional SAGE</strong> (Shapley Additive Global importancE) values for any black-box predictor, with asymptotically valid Wald confidence intervals and hypothesis tests for global feature importance.</p> <p>The conditional value function E[f(X) | X_S = x_S] is estimated via a <strong>Gaussian copula with empirical marginals</strong> (default) or a <strong>nonparametric vine copula</strong> (<code>rvinecopulib</code>). A one-step debiased estimator with K-fold cross-fitting and U-statistic debiasing of the squared loss yields root-n consistent, asymptotically normal SAGE estimates (Theorem 1 in the paper).</p> <h3>Main functions</h3> <p>| Function | Description | |----------|-------------| | <code>sage_cv_gcop()</code> | Conditional SAGE via Gaussian copula — <strong>recommended</strong> | | <code>sage_cv()</code> | Conditional SAGE via nonparametric vine copula | | <code>sage_cv_marginal()</code> | Marginal (interventional) SAGE — baseline | | <code>fit_gauss_copula()</code> | Gaussian copula fitting with empirical marginals | | <code>fit_copula()</code> | Nonparametric vine copula fitting (<code>rvinecopulib</code>) |</p> <h3>Installation</h3> <pre><code class="language-r">remotes::install_github("agostinognasso/shapCopula") Requires: R ≥ 4.1.0, rvinecopulib ≥ 0.6.0 --- </code></pre> |
| title | agostinognasso/shapCopula: shapCopula v0.1.0 — Initial release |
| url | https://doi.org/10.5281/zenodo.20392904 |