- Added for
mixstanvar
automatic generation of distribution functions allowing truncated priors inbrms
with mixtures - Slight speed increase for Stan model by more efficient construction
of likelihood. Also reducing object size of
gMAP
objects by avoiding to store redundant variables. - Upgraded
testthat
edition to version 3.
- Fix issue #18 of ESS ELIR aborting whenever one mixture component has zero weight.
- Fixed a rare issue when estimating ESS ELIR for beta mixtures. The calculation aborted due to boundary issues which are now avoided.
- Avoid over and underflow of beta-binomial distribution.
- Replace calls to deprecated
ggplot2::qplot
with respective calls toggplot2::ggplot
. - Use new
array
notation formixstanvar
generated Stan code to make generated Stan programs compatible with Stan >= 2.33
- Updated Stan model file syntax to use new array syntax as required by Stan >=2.33. This upgrades the minimal Stan version to 2.26.
- Moved most vignettes to be articles to decrease size of R packages. Articles are available on the package homepage.
- Added
pngquant
to system requirements of package as requested by CRAN.
- Add unit tests for
mixstanvar
function and disabled integration tests on CI/CD systems as these require significant resources due to the need to compile Stan models. The unit tests do ensure that things are correct from theRBesT
side of things. - Use new scheme of
roxygen2
to document package documentation.
- Make EM for normal and multivariate normal mixtures deterministic in that the resulting density does not depend on the random seed any longer, which is the intended behavior. Note that this change may cause that estimated normal / MVN mixtures will be numerically different as compared to the previous version. The estimated density itself is (statistically) still the very same.
- Allow use of mixture priors in
brms
models with the new adapter functionmixstanvar
. - Allow use of named dimensions for multivariate normal mixtures.
- Add experimental diagnostic plots for mixture multivariate normal EM fits. These are subject to changes in future release and user feedback is very welcome.
- Compress png plots of vignettes saving ~40% in file size of the package sources.
- Fix issue when plotting EM diagnostic debugging plots for normal mixtures.
- Comply with newer ggplot2 standards to not use
aes_string
.
- Implementation multivariate normal mixture support in a first version. This includes density evaluation, basic summary functions and multivariate normal EM fitting. Support is not yet as complete as for other densites, but will be expanded in upcoming releases.
- Change the default for the option
constrain_gt1
of the EM for beta mixtures toTRUE
. This will be default constrain the fitted parameters of each beta mixture component to be greater than unity as required for finite ESS elir calculations.
- resolve compilation from source issues on some platforms triggered
by changes in
rstantools
- correct documentation on difference distribution and improve PoS with co-data vignette
- ensure C++17 compatiblity per CRAN (triggers an issue with clang 16)
- fix links in README.md to now link to new public pkgdown web-site
- upon package load
RBesT
will now report the date of the release and the respective git commit hash used to create the sources of the package.
- ensure that
predict
for new studies will sample the study specifc random effect per iteration only once. This is important for MAP priors with covariate effects (which are pooled over the studies).
- use clustermq inplace of batchtools for SBC runs. Also use L'Ecuyer CMG random number gen during SBC runs.
- expand introductory vignette with plot of ESS vs robust weight
- start using
matrixStats
to speed up EM algorithms & OCs - avoid warning whenever 2S normal design expanded their domain when called repeatedly
- add
mixdist
plot when plotting a mixture resulting from *mixfit call - add warning message when printing a
gMAP
analysis for which the Stan sampler had issues due to divergent transitions or non-convergence as indicated by large Rhat
- mixture density evaluations (
dmix
) with a defined link function did not evaluate correctly, which was visible when plotting mixtures (with more than one component) with link functions
- update references to JSS publication
- link against RcppParallel to align with new Stan requirements
- address CRAN comments
- stabilize elir ESS integration by integrating per mixture component
- comply with forthcoming and stricter stanc3 Stan transpiler
- address some warnings from ggplot2 3.3.0
- fix CI system issues
- fix issues with normal decision2S_boundary when boundaries are grown
- add demo for 2S OC simulation code for time-to-event endpoint with constant hazard assumption
- drop tidyverse dependency
- expand SBC checks to include group specifc estimates
- stop setting the ggplot2 default theme when loading package. All plots now use the bayesplot theme which can be modified with bayesplot_theme_* functions. See ?bayesplot::bayesplot_theme_get.
- correct transformation issue in MAP for variances vignette - thanks to Ping Chen
- allow for constrained fitting of beta mixtures which have a & b parameters greater than 1. This is the new default behavior which the function will inform about. The informational message will be removed in a future release.
- introduced new mixecdf plot as diagnostic for EM fits
- Now really fix n2=0 case for 2S design functions for indirect comparisons
- Update package structure to new rstantools 2.0 system
- Fix vignette MAP for variances (missing definition)
- Speedup example run time
- Avoid use of cat in functions and use message instead
- Replace dontrun in examples with donttest
- Require NOT_CRAN=true for tests to run
- Work around compiler warning with clang on fedora platform
- Fix indirect comparisons to work with normal/Poisson/binomial (inexact) to allow for n2=0 in oc2S calls.
- Make mixture quantile function more robust to work with very flat mixture priors.
- Align ESS Morita calculations with Neuenschwander B. et al., pre-print 2019; arXiv:1907.04185
- Introduce elir ESS method as new default for ESS
- Allow to sample prior predictive with gMAP (argument prior_PD)
- Switch internally to ab parametrized version of EM beta algorithm
- Use Simulation Based Calibration for gMAP model qualification
- Improve covariate handling (naming of data items)
- Speedup Stan model by avoiding matrix-vector products with many zeros
- Fix index issue with differential discounting when used with covariates
- Make initialization of EM algorithms more robust
- Avoid special build hacks on MacOS
- Address issue for build process on MacOS.
- Re-create vignettes with proper MCMC sampling.
- Automate R package build process using CI/CD.
- Corrected 1.3-4 release notes to include MAP for variances vignette
- Make build process more robust (updated src/Makevars{.win})
- Added probability of success at interim vignette
- Added probability of success with co-data vignette
- Make package work with rstan 2.18.1.
- Revert BetaBinomial implementation back to R functions.
- Bugfix for decision1S_boundary for normal case for extreme parameter configurations (fixes pos1S & oc1S as well).
- Bugfix for mixcombine and plot with normal mixtures without a sigma being defined.
- Bugfix for repeated calls to decision2S_boundary for normal endpoint (fixes pos2S & oc2S as well).
- Avoid use of deprecated bayesplot function arguments whenever divergencies occured.
- (corrected) Added MAP for variances vignette
- Change numerical equality testing to use expect_equal (which uses all.equal internally accounting for machine specifc tolerances) to pass tests for no long double case. Numerical tolerances are reverted back to 1.3-1 settings.
- Adjust numerical tolerances to pass tests for no long double case
- Add Trustees of Columbia copyright for respective files in DESCRIPTION
-
Added probability of success calculation for 1+2 sample case.
-
Added decision1+2S_boundary functions (and deprecated use of y2 argument of oc functions)
-
Added RBesT.integrate_args option for greater control over density integrations.
-
Correct cumulative predictive of beta mixtures to return 0/1 for out-of-range values (instead of leaving those out).
-
Deprecated functions oc1+2Sdecision which are replaced by decision1+2S.
- Fix plotting procedures to work with bayesplot 1.3.0
- Further speedup example runtimes.
- Compactify reference PDF manual.
- Introduce sampling arguments to gMAP.
- Shorten runtime of examples.
- First CRAN release.
- Update of documentation.
- Redesign of reference scale handling for normal case.
- Enable standard error as sufficient statistic in \code{postmix} function.
- Introduced plotting options.
- Increased adapt_delta default and set stepsize+max_treedepth default.
- Added RBesT.MC.{ncp, init, rescale} option.
- Corrections for Poisson OC.
- pmixdiff function now integrates over full support.
- Added \code{link} argument to \code{oc2Sdecision} which enables designs based on log-odds decisions or relative risks.
- New graphical vignette and new forest plot function.
- Use \pkg{bayesplot} as standard plotting package.
- Stabilize integration in pmixdiff for beta mixtures by logit transform.
- Set default adapt_delta to 0.975.
- Made RBesT compatible with ggplot2 2.0.
- Allowed n2=0 in \code{oc2S} function.
- Corrected Poisson stratified estimates.
- Added warning on divergent transitions.
- Added the crohn dataset.
- Minor typo fixes.
- First release.