NEWS

---

selSurv 0.1.1 (2026-10-31)

Initial CRAN release.

• Updated the simulation-study reference to the published journal DOI.
• Clarified public method descriptions and documentation.
• Simplified the README while retaining development smoke tests.

selSurv 0.1.0.9016

• Clarified the supported penalty scope of the inferential workflows. coxlasso_fit() continues to expose the glmnet elastic-net mixing parameter for fitting flexibility, while the inferential validation in selSurv currently concerns Lasso screening (alpha = 1).
• Restricted coxlasso_debias() and the associated wild and Efron score-bootstrap backends to Cox-Lasso pilots with alpha = 1; non-Lasso pilots now fail explicitly rather than returning apparently validated debiased inference.
• Kept split-sample, fixed-support model-based bootstrap, and procedure-level re-selection workflows computationally available for non-Lasso mixing parameters, but added explicit warnings that these settings have not been separately validated inferentially in selSurv.
• Added a package workflow guide documenting inferential targets, bootstrap interfaces, penalty-parameter naming, penalty scope, and tie-handling distinctions.
• Added focused regression tests for Lasso-only debiasing safeguards and non-Lasso warning behavior.

selSurv 0.1.0.9015

• Made Cox tie conventions explicit across the model-based bootstrap workflows: penalized Cox-Lasso selection uses the Breslow convention, ordinary post-selection Cox refits use Efron's method, and model-based event generation uses an explicit Breslow cumulative baseline hazard.
• Added compatibility handling for installed glmnet versions that do not formally support cox.ties, avoiding silently ignored tie arguments while preserving the historical Breslow selection convention.
• Made the Cox-Lasso cross-validation convention explicit as deviance loss with grouped cross-validation, while retaining CV-min, 10 folds, and Lasso mixing alpha = 1 as the general fitting defaults.
• Updated the general re-selection defaults to conf_level = 0.95, seed = NULL, and standardize = TRUE. A randomly generated root seed is stored when seed = NULL, after which bootstrap-generation and CV streams remain deterministic and method-order invariant.
• Retained rng_mode = "legacy_nki70" and its historical Breslow, random-stream, solver-tolerance, and standardization conventions solely for exact reproduction of the archived NKI70 analysis.
• Revised coxlasso_split() to separate interval confidence through conf_level = 0.95 from Cox-Lasso mixing through penalty_alpha = 1. Historical alpha calls remain supported as a deprecated confidence-level bridge.
• Kept the historical direct bootstrap backend defaults unchanged for compatibility; the unified coxlasso_boot() interface remains the recommended bootstrap entry point with B = 500, conf_level = 0.95, and seed = NULL.
• Added regression tests for tie handling, Breslow baseline generation, cross-validation conventions, re-selection seed semantics, general-purpose defaults, and split-sample compatibility.

selSurv 0.1.0.9014

• Standardized categorical data-frame inputs to stable treatment coding: factors, ordered factors, character variables, and logical variables now contribute (k-1) design-matrix coordinates using the first level as the reference, independent of data-frame column order.
• Ordered factors are treated as ordinary categorical predictors rather than using polynomial contrasts; logical predictors use FALSE as the reference level. Numeric and integer covariates remain numeric.
• Character predictors are converted to unordered factors. Users who require a particular reference category should supply a factor with an explicit level order or use relevel() before fitting.
• Selection and re-selection continue to operate on individual design-matrix coordinates, not on whole categorical variables as groups. Consequently, Lasso selection can depend on the chosen reference level.
• Added explicit checks for constant categorical predictors, unsupported data-frame column classes, empty inputs, missing or non-finite values, invalid or duplicated design-column names, and generated dummy-name collisions.
• Added design-coordinate checks so that post-selection refits and externally supplied original workflows cannot silently use reordered or differently coded covariates.
• Expanded regression tests for factor, ordered-factor, character, logical, numeric, sparse-matrix, reference-level, contrast-option, input-boundary, and procedure-level re-selection behavior.

selSurv 0.1.0.9013

• Centralized covariate design-matrix normalization across the main Cox-Lasso workflow and procedure-level re-selection.
• coxlasso_boot_reselect() now accepts data-frame covariates using the same model.matrix(~ . - 1) expansion already used by coxlasso_fit(), including factor and logical columns.
• Added explicit rejection of missing and non-finite covariate values before model fitting.
• Added regression tests for numeric, factor/logical data-frame, and sparse-matrix inputs.
• Preserved the existing coding convention, numerical behavior, and archived NKI70 re-selection parity.

selSurv 0.1.0.9012

• Added an experimental procedure-level model-based bootstrap with Cox-Lasso re-selection through coxlasso_boot_reselect().
• Added fixed-support, fixed-penalty re-selection, bootstrap CV-min, CV-1SE, and AIC tuning variants.
• Non-selected coordinates are recorded as zero; failed non-empty unpenalized Cox refits remain missing rather than being converted to zero.
• Added method-level diagnostics for support overlap, model size, selection/refit success, tuning parameters, and coordinate-level re-selection frequencies.
• Added method-order-invariant random streams for normal use.
• Added rng_mode = "legacy_nki70" solely for reproducibility of the archived NKI70 analysis. The B=20 regression gate reproduces the archived support diagnostics, including the historical AIC refit failure pattern.
• The existing coxlasso_boot(inference = "conditional") semantics remain fixed-support and are not broadened to include procedure-level re-selection.

selSurv 0.1.0.9011

• Integrated the August 2026 unified coxlasso_boot() API with the recovered v12 package source.
• Added dispatch between debiased wild bootstrap and Efron's bootstrap and fixed-support model-based bootstrap.
• Added compatibility with the established v12 selSurv_debias class.
• Kept procedure-level re-selection outside the first unified bootstrap API; re-selection remains available through coxlasso_boot_parametric().
• Added validation/dispatch tests and development use-case documentation.
• Archived the historical selSurv IWSM text as an unpublished draft and noted that the retained IWSM 2026 proceedings contribution was the PoSI work.
• Removed local .RData/.Rhistory workspace artifacts from the source tree.

selSurv 0.1.0.9001

• Added user-facing coef and se fields to coxlasso_fit() and coxlasso_refit() outputs.
• Added support-specific coef_support and se_support fields to coxlasso_refit().

selSurv 0.1.0.9002

• Added documentation for S3 print methods.
• Rewrote workflow Rd usage/argument documentation.
• Moved local smoke-test examples and implementation notes under inst/.
• Reduced R CMD check warnings/notes caused by package documentation layout.

selSurv 0.1.0.9003

• Fixed Rd code/documentation mismatches for coxlasso_wildboot() and coxlasso_efron_events().
• Added .Rbuildignore entries for stale top-level example/development files.
• Shortened the coxlasso_fit() Rd usage block.

selSurv 0.1.0.9004

• Added explicit roxygen usage block for coxlasso_fit() to remove Rd line-width NOTE.
• Broke the coxlasso_fit() function signature across lines.
• Removed empty inst/extdata from the source tree.

selSurv 0.1.0.9005

• Removed the obsolete combined coxlasso_workflows.Rd file to avoid duplicate Rd aliases after roxygen generates individual workflow documentation.
• Removed the manual coxlasso_fit.Rd file so roxygen can regenerate it from the explicit multi-line @usage block.

selSurv 0.1.0.9006

• Added plot_selection_diagnostics() for base-R visualization of re-selection and zero-frequency diagnostics.
• Added tests for diagnostic comparison and plotting helpers.

selSurv 0.1.0.9007

• Improved plot_selection_diagnostics() layout; default legend is now a compact top legend and neutral greys are used by default.

selSurv 0.1.0.9008

• Added a prettier default Cleveland-style dot plot to plot_selection_diagnostics().
• Kept grouped bars available via plot_type = "bar".
• Improved plot readability for many variables and small RStudio plot panes.

selSurv 0.1.0.9009

• Added a colorblind-friendly color palette to plot_selection_diagnostics().
• Added a robust legend_position = "side" default using a separate legend panel.
• Kept grouped bars available and improved the color bar plot option.

selSurv 0.1.0.9010

• Replaced the default side legend by a top legend panel in plot_selection_diagnostics().
• This avoids legend clipping in the RStudio zoom window and keeps the main plot area readable.

selSurv 0.1.0.9000

Initial research implementation.

• Added Cox-Lasso fitting workflow.
• Added support-restricted Cox refitting.
• Added approximate debiasing workflow.
• Added multiplier wild and event-level score bootstraps.
• Added model-based bootstrap with fixed-support, fixed-lambda re-selection, and retuned re-selection modes.
• Added split-sample workflow.
• Added target and method summary helpers.
• Added vignette explaining target definitions and workflow interpretation.

selSurv 0.1.0

Initial public GitHub release.

• Provides Cox-Lasso screening, support-restricted Cox refitting, split-sample inference, approximately debiased coefficient estimates, and score-level and model-based bootstrap workflows.
• Supports fixed-support inference and procedure-level re-selection, with explicit distinctions between the corresponding inferential targets.
• Corrects the Cox score and observed information calculations for tied event times under the Breslow and Efron conventions.
• Includes numerical regression tests against an independent Cox partial-likelihood reference implementation.
• Documents the supported Lasso penalty scope and the conventions used for fitting, refitting, and bootstrap inference.