NEWS
spsurv 1.1.0 (2026-09-28)
- CRAN resubmission: Maintainer contact updated to
[email protected];
license is GPL-3 (removed the bundled LICENSE copy of the GPL);
Authors@R roles reviewed (single maintainer; thesis advisors listed as
ths, with rev for code review feedback); package Description citation
updated to <doi:10.48550/arXiv.2003.10548>; set
Config/rstantools/auto_config: FALSE and ship committed Stan C++ exports
(no install-time configure/rstan_config(), avoiding Windows R-devel
rstan.dll load failures); added RcppParallel and synced
src/Makevars.win TBB flags with Unix Makevars; restored covr,
rsample, and rsurv in Suggests for optional tests referenced in
tests/.
- Inference robustness:
.spbp_gamma_information_diagnostics() no longer
calls qr() on non-finite gamma information matrices; vcov()/survfit()
return NA uncertainty with a warning when that block is unavailable.
- tidymodels: parsnip engines
spsurv (MLE) and spsurv_bayes for proportional_hazards(), new proportional_odds(), and survival_reg(); bp_survival_reg() convenience constructor; workflow / predict(type = "survival") support.
- tidybayes:
as_draws_df.spbp(), spread_surv_draws.spbp(), and S3 hooks for tidy_draws / spread_draws / gather_draws on Bayes fits; posterior draw indices (.chain, .iteration, .draw).
- Prediction: censored-style
predict.spbp() types (survival, time, linear_pred); augment.spbp().
- Vignette:
vignette("tidymodels").
- Inference:
vcov.spbp() no longer applies a diagonal ridge (polish removed); beta standard errors are unchanged when the Bernstein block is stable and no longer inflated when it is ill-conditioned.
- Summary:
summary.spbp() now includes a coef_interval component (Wald for MLE, HPD for Bayes) and prints it when show_intervals = TRUE.
- Repository: paper workflow moved from
inst/ to paper/ (excluded from the package tarball).
- Dependencies: trimmed
Suggests to packages used in vignettes, tests, or optional integrations. devtools and roxygen2 are not listed in Suggests.
spsurv 1.0.2
- Vignettes: Expanded with EDA plots, visual checks, and decision guidance across all seven articles.
- tidy/broom:
tidy(), glance(), logLik(), AIC(), anova(), estimates(), and se() methods for spbp objects; bernstein() baseline specification.
- Prediction:
predict() and improved survfit() (tidy output, newdata covariates, Bayesian monotone bands).
- Residuals:
residuals() supports martingale, deviance, and Cox-Snell types; martingale diagnostics fixed in examples.
- Handlers: Removed
handlers.R; validation and setup logic inlined in spbp.default.
- vcov: Block-wise inversion of the Hessian (regression vs Bernstein polynomial blocks) in
vcov.spbp.
- Stan: Slim
transformed parameters (only alpha stored); pointwise log_lik moved to generated quantities; likelihood via bp_pointwise_log_lik() in the model block.
- Stan (AFT): No clamping of
u to avoid boundary bias; minimum feasible divisor (min_range) only when range is degenerate.
- MLE: Initial values for the optimizer drawn from prior distributions; retries limited to 3 attempts.
spbp objects: Component degree on fits and in call$degree; default Bernstein degree ceiling(sqrt(n)) when omitted (documented in help, README, and vignettes).anova.spbp: Pairwise and sequential analysis-of-deviance tables aligned with survival::survreg (Terms, Test, Df, Deviance, etc.).bpph / bppo / bpaft: User-facing match.call() without unevaluated dist/baseline symbols; optional degree argument.- Documentation: Expanded
spbp / spbp.default help (@details, return components, vignette links); README usage and model-comparison examples updated.
- Examples:
data("veteran", package = "survival") in docs and examples.
- pkgdown: Reference index updated to match current package; Bootstrap 5 / default template; logo size and alt-text; README aligned with site style.
- Repository: CRAN submission hygiene (
.gitignore, .Rbuildignore, LICENSE).
spsurv 1.0.1
spsurv 1.0.0 (2020-03-31)
- This is the first release of spsurv.