- upon package load
`OncoBayes2`

will now report the date of the release and the respective git commit hash used to create the sources of the package.

- ensure compatibility with posterior 1.4.0 (posterior returns
`num()`

formatted columns, which`OncoBayes2`

is not supporting) - ensure C++17 compatiblity per CRAN (triggers an issue with clang 16)

- add warning messages when printing
`blrmfit`

objects for divergent transitions or non-convergence of parameters - add diagnostic extraction functions
`nuts_params`

,`rhat`

,`log_posterior`

and`neff_ratio`

for`blrmfit`

objects as defined in`bayesplot`

- speedup
`pp_data`

by about 25% which is used in all posterior summary functions - add
`ewoc_check`

summary routine for`blrm_trial`

objects. The returned summary contains MCMC diagnostics and accuracy estimates of the EWOC metric as defined for the trial. This allows to assess the MCMC estimation error for EWOC. - added warning for imprecise estimates of the EWOC metric whenever
`blrm_trial`

objects are printed.

- fix broken exclusion of unnecessary random variables whenever
`save_warmup=FALSE`

.

- allow cmdstanr as new backend for blrm_exnex. The Stan model file is written with the function cmdstanr::write_stan_file to disk such that the model binaries can be cached if the user defines the global option cmdstanr_write_stan_file_dir. See ?cmdstanr::write_stan_file (requires cmdstanr >= 0.5.0)

- fix issue with example_model not exposing properly objects when running with silent=TRUE

Switch SBC runs to use clustermq in lieu of batchtools. Also now use L’Ecuyer CMG as rng engine during SBC runs.

add critical_quantile function allowing to calculate critical doses which fulfill conditions like EWOC.

- Make Stan model compile with 2.27+ (lupmf postfix for variables is not allowed)

- Drop unneccessary random variables from the posterior whenever save_warmup=FALSE. This decreases the size of the posterior in memory by ~60% for the combo2 example.

- Address CRAN comments

- Significantly speedup Stan model by dropping normalization of binomial
- The default interaction model for blrm_trial is now saturating (blrm_formula_saturating) in logit space, improving behavior for combinations with antagonistic toxicity.
- blrm_trial simplified prior now defaults to ENXEX off, and prior parameter defaults have been updated.
- New save_warmup argument to blrm_exnex can be used to disable saving of warmup samples, which defaults to the OncoBayes2.MC.save_warmup option, which defaults to TRUE in line with previous versions. NOTE: Starting with the next version 0.9.0 the new default will be FALSE!
- Do not crash for summary(trial, “data_prediction”) if blrm_trial has NULL or 0-row data
- Spurious warning for dose_id NA on blrm_trial instantiation removed
- Spurious warning for dose not prespecified on summary(…, newdata=…) removed
- New plot function plot_toxicity_intervals_stacked(), which visualizes model posterior or predictive probabilities in discrete intervals, over a continuous range of doses
- Automated visual regression testing implemented using the vdiffr package
- Fix in plot_toxicity_curve to avoid overplotting if newdata contains columns other than grouping variables and doses (e.g. dose_id)
- Avoid re-parsing model formula whenever newdata is not used in respective functions. This prevents that changes to global variables (like dref) have any effect whenever the fitted data only is used.

- Catch data specification errors for wrongly nested groups/strata
- BLRM trials now also print toxicity probability intervals and the EWOC setting
- The blrm_formula_linear and blrm_formula_saturating interaction model formula generators can now be customized to generate specific interactions using the specific_interaction_terms argument.
- Add predictive summaries to summary method.
- Switch predictive outputs to be based on averaging sampling density. This avoids the need for additional sampling for predictive summaries.
- Add plotting methods for blrmfit and blrm_trial objects.

- Drop RBesT dependency
- Avoid setting the ggplot2 theme upon package load
- Enhance SBC runs to monitor and report sampler diagnostics

- Add Operation Qualification script (run-oq.R in inst/extra)
- Add saturating interaction model (blrm_formula_saturating) and associated tests

- 2nd attempt to work around issues found by clang sanitizer in Stan model.

- Work around issues found by clang sanitizer in Stan model.

- Fix issue with upcoming R 4.0 which changes stringsAsFactors default to FALSE
- Fix for upcoming tibble 3.0 upgrade which changes conventions of growing factors within tibbles

- Fix issue with model outputs when EXNEX is being used. We recommend all users to upgrade to this version. The SBC runs now include checks for the per-group estimates to avoid a regression of the bug.
- Correct printing of prior information wrt. to summaries of number of strata and groups.

- Merge data with same group and dosing in blrm_trial before passing into blrm_exnex for improved performance
- Ensure consistent sorting of this data to improve reproducibility of output with respect to input data permutations
- speedup Stan model which now skips data rows with no cases
- correct blrm_exnex documentation to reflect correctly exchangability model used for the interaction model

- Improve numerical stability of log_inv_logit function, preventing NaNs in the output of pp_data and resulting errors in summary()

- Support summary.blrm_trial(…) syntax passing into summary.blrmfit()

- Fixed Roxygen for posterior_predict

- fix vignette documentation
- correct blrm trial co-data example
- fix test issues per CRAN checks on macosx-old-r run

- new blrm_trial function which facilitates dose-escalation trial conduct by combining key trial design features
- new add_data argument to update function which adds data to existing model objects of class blrmfit or blrm_trial
- various smaller bug fixes

- run in all examples the example code (remove dontrun sections)

- use message instead of cat in functions using printing except summary or print
- suppress by default messages from Stan, can be enable with verbose=TRUE
- make examples run with very short sampling

- Correct loading and exporting of methods
- Add tidybayes example for continuous use and visutalization of model
- Allow multiple central probability widths in prob argument of summary method

- Correct external package loading in examples

- Added function prior_summary
- Restructured print output
- Added examples single-agent, combo2 and combo3 along with example data sets
- A lot more documentation on reference pages
- New vignette on standard use case of blrm_exnex in Oncology
- Qualified blrm_exnex model with Simulation Based Calibration

- Initial release