This is an update of mlumr from 0.1.0, which is on CRAN, to 0.2.0. NEWS.md
lists the changes. The largest are time-to-event outcomes (parametric and
M-spline baselines, fitted to reconstructed Kaplan-Meier data) and prediction
of marginal effects in arbitrary target populations.
Local R CMD check --as-cran --no-manual mlumr_0.2.0.tar.gz on macOS 27.0.1
(aarch64-apple-darwin23), R 4.6.0:
- 0 errors
- 1 warning, environmental: the
checkbashismsscript is not installed on this machine, so the check reports that "a complete check needs the 'checkbashisms' script." CRAN's Linux servers have it. - 0 notes. The CRAN incoming feasibility check reports, as information, that
cmdstanris in Suggests and is not in a mainstream repository. It is available from the declaredAdditional_repositories: https://mc-stan.org/r-packages.
Tests under --as-cran: [ FAIL 0 | WARN 32 | SKIP 93 | PASS 2853 ]. The 93
skips are the Stan-sampling tests, which are marked skip_on_cran() because
each one compiles or samples a model. They run in a separate CI job with
NOT_CRAN=true (below), where they pass.
- local macOS 27.0.1 (aarch64-apple-darwin23), R 4.6.0
- GitHub Actions,
R CMD check --no-manual --as-cran, R release: macOS, Windows and Ubuntu. Status OK on all three. - GitHub Actions, Ubuntu, R release, full test suite with the Stan tests enabled under CmdStan: 4705 passed, 0 failed, 0 skipped.
None. No CRAN package depends on, imports, suggests or links to mlumr.
- The package embeds ten Stan models: binomial, normal and Poisson, each with SPFA and relaxed variants, plus parametric and M-spline survival in both variants. They are compiled at install time through the standard rstantools pipeline, so installation takes several minutes of C++ compilation.
- The installed size is 14.9 Mb:
libs10.2 Mb (the compiled Stan models),R2.1 Mb anddoc2.0 Mb. As with other Stan-backed packages such asrstanarmandbrms, the shared library dominates. The source tarball is 1.9 Mb. cmdstanris an optional backend. rstan is the default engine and is on CRAN. Examples and tests do not require cmdstanr, and the cmdstanr comparison code is skipped unless cmdstanr and CmdStan are available.- Seven of the nine vignettes fit Stan models, so they are pre-rendered and
shipped through the
R.rsp::asisengine; re-building them runs no Stan. The other two are knitr vignettes:introductionevaluates no code, anddata-preparationruns data-setup code only, with no model fitting.