- provide optional (
optimalWarmStart = FALSE
) more parallelization when profiling likelihood - make
maxResets
a function parameter - ensure compatibility with R v4.4
Changes:
- fix some NaN/-Inf inconsistencies with likelihood profiling
Changes:
- remove dependence on
BH
- improvements on adaptive likelihood profiling
- add
auto
option tocvRepetitions
- bumped explicit C++11 requirement up to R v4.1
- removed deprecated use of
dbplyr:::$.tbl_lazy
a. breaking change indbplyr v2.4.0
Changes:
- fix uninitialized value in detected in
computeAsymptoticPrecisionMatrix()
; value was priorType - fix memory leak caused by call to
::Rf_error()
- fix line-endings on Makevar on windows
- bump for R 4.2
- fix CRAN warnings
a. used
minValues
- fix CRAN notes a. remove explicit dependence on C++11 (except for R <= 4.0)
Changes:
- fix small memory leak caused by direct call to '::Rf_error()'
- disable JVM calls on CRAN due to uninitialized memory in Java JVM
Changes:
- fixed likelihood profiling when non-convex due to numerical instability
- fixed parsing of 64-bit covariate IDs
- fix BLR convergence criterion when there are 0 (survival) events
- add Jeffrey's prior for single regression coefficient
Changes:
- adaptive likelihood profiling when objective function is concave.
Changes:
- specify new option
nocenter
insurvival
package when testing predicted hazard function
Changes:
- implement Fine-Gray competing risks regression
- fixed
getCyclopsProfileLogLikelihood
when starting with extreme coefficients
Changes:
- switch to
Andromeda
fromff
to hold large datasets. This change breaks API
Changes:
- removed, unused variable imputation functions that contained a
std::exit
Changes:
- fix computation under conditional Poisson models by reverting to v1.3.4-style loop
- fix several unit-tests for compatibility with
R 4.0
factors - add ability to profile likelihood function in parallel
- add initial infrastructure for competing risks models
Changes:
- use
RNGversion("3.5.0")
in unit-tests to reproduce old RNG behavior - fix prior-type checks when specifying multiple types
Changes:
- patch two memory leaks in
ModelData.cpp
andModelSpecifics.hpp
Changes:
- simplify internal transformation-reductions loops
- implemented non-negative weights
- allow for run-time selection of 32-/64-bit reals
- remove dependence on GNUmake
- temporarily remove dependence on
RcppParallel
(until TBB is again R-compliant)
Changes:
- fix undeclared dependencies in unit-tests:
MASS
andmicrobenchmarks
- fix issues with ATLAS compilation
- add contexts to testthat files
- fix ASAN errors in
AbstractModelSpecifics
Changes:
- fix testthat expected error message
Changes:
- explicitly includes
<complex>
header, needed forR
3.5 builds - remove
pragma
statements used to quietRcppEigen
andRcppParallel
Changes:
- fixes covariate indices returned from
.checkCovariates
when excluding covariates from regularization
Changes:
- implements specialized priors through callbacks for use, for example, in the BrokenAdaptiveRidge package to provide L0-based model selection
- implements specialized control through callbacks for use, for example, auto-and-grid-based cross-validation hyperparameter searches
- removes src/boost that clashes with BH 1.65.0
Changes:
- fixed
predict
error withff.data.frame
with size == 0
Changes:
- fixed
solaris
build errors - added compatibility for C++14 (make_unique)
- fixed multiple ASan warnings
Changes: initial submission to CRAN