islasso 1.6.1 (2025-11-13)

Created GitHub page and website.
Created introductory vignette.
Fortran routines updated and speeded up (~2x).
Manuals updated.
Added new function relax.islasso(), which allows fitting a relaxed islasso model by selecting variables to remain unpenalized. Variables can be specified either by name or by index, or automatically selected according to a significance level (alpha). This extension provides additional flexibility in post-selection inference.
Some bugs fixed.

islasso 1.6.0 (2025-07-30)

Core computational routines have been cleaned up, and some bugs have been fixed.
Legacy R routines have been revised, cleaned, and commented. Minor inconsistencies have been addressed.

Help files and function manuals are now fully managed via roxygen2, with substantial updates to usage examples and descriptions.

All plotting functions have been refactored to use the ggplot2 framework for consistent and modern graphics.

A custom ASCII startup banner has been added on package attach, providing a welcoming and informative message.

New optimization algorithm for the ‘islasso’ method. The algorithm is now stable for all the implemented distributions.
In aic.islasso() function the available methods are “AIC”, “BIC”, “AICc”, “eBIC”, “GCV”, “GIC”.
New class of functions named islasso.path created. The main function islasso.path() builds the coefficient profile for a fixed sequence of lambda values.
New function GoF.islasso.path() extracts the optimal tuning parameter minimizing a fixed criterion. Available criteria are the same as in aic.islasso().
Some bugs fixed.

New implementation of the estimating algorithm. Now islasso is much stabler and faster.
New function: general linear hypotheses for linear combinations of the regression coefficients, including confidence intervals.
Prediction function includes confidence intervals for the fitted values.
Step halving with Armijo’s rule improved.
Convergence criterion improved.
Some bugs fixed.

New implementation of the estimating algorithm. Now islasso is much stabler and faster, reducing the number of iterations to reach convergence.
Step halving with Armijo’s rule implemented.
Elastic-net approach added via alpha parameter in the objective function (as in glmnet).
Summary method now includes degrees of freedom for each covariate, with choice between t-test or z-test (only for Gaussian family).
optim.islasso renamed to aic.islasso; interval specification no longer required.
islasso.control renamed to is.control; control parameters modified.
Two trace versions implemented in is.control: compact (trace = 1) and verbose (trace = 2).
Some bugs fixed.