Added support to compute and compare growth curve by using marginaleffects package as back-end.
(see marginal_draws(), marginal_comparison() and growthparameters_comparison()).
This allows utilizing the computation flexibility offered by the marginaleffects package to
estimate various quantities of interest such as adjusted growth curves (distance and velocity) and
growth parameters such as age at peak growth velocity. All three functions allow parallel computation
via future and doFuture packages.
An experimental support for using $pathfinder() based initial values for the MCMC sampling
$sample() (via argument pathfinder_init = TRUE, default FALSE). The arguments for the
$pathfinder() can be specified as a named list via the pathfinder_args. Note that this
feature is only available when backend = 'cmdstanr'.
The default distribution for all parameters i.e., regression coefficients as well as the standard
deviation (sd) for the group level random effects and the distributional parameter (sigma) changed
to the normal(). Earlier, the distribution for regression coefficients and the sd for the
group level random effects was student_t() whereas distribution for sd of distributional
parameter (sigma) was exponential(). Note that the same location and scale parameter for the
student_t() which were used earlier are now used for the normal() distribution. Similarly,
the scale parameter used earlier for the exponential() for distributional parameter is now
used for setting the normal() prior for the distributional parameter (location parameter as '0'.
The default initials for all parameters i.e., regression coefficients as well as the standard
deviation (sd) for the group level random effects and the distributional parameter (sigma) changed
to the random.
envir = globalenv() for post processing functions. The environment is now automatically set to match the environment of exposed functions. It is important to note that setting environment manually (via the 'envir' argument) may actually result in errors. The 'envir' argument is now mostly for internal use only which is needed during tests.add_model_criterion() function to compute fit criteria such as "loo", "waic", "kfold",
"loo_subsample", "bayes_R2" (Bayesian R-squared), "loo_R2" (LOO-adjusted R-squared), and "marglik"
(log marginal likelihood). The computed fit criteria are added to the model object for later use including
comparison of models. The add_model_criterion() is a wrapper around the add_criterion()
function available from the brms packagebsitar::bsitar()```received options ```file```, file_refit, and ``file_compress to save and retreive fitted objects. See brms::brm help file for details.