Tidybayes supports two classes of models and sample formats: Models/formats that provide prediction functions, and those that do not.
All supported models/formats support the base tidybayes sample extraction functions, such as
tidy_draws(), spread_draws(), and gather_draws(). These models/formats include:
rstan models
brms::brm() models
rstanarm models
runjags::runjags() models
rjags::jags.model() models, if sampled using rjags::coda.samples()
jagsUI::jags() models
MCMCglmm::MCMCglmm() models
coda::mcmc() and coda::mcmc.list() objects, which are output by several model
types.
Any object with an implementation of coda::as.mcmc.list(). For a list of those available in your
environment, run methods(as.mcmc.list)
If you install the tidybayes.rethinking package (available at
https://github.com/mjskay/tidybayes.rethinking), map and
map2stan models from the rethinking package are also supported.
In addition, the following models support fit and prediction extraction functions, such as
add_fitted_draws() and add_predicted_draws():
brms::brm() models
rstanarm models
If your model type is not in the above list, you may still be able to use the add_draws()
function to turn matrices of predictive draws (or fit draws) into tidy data frames.
If you install the tidybayes.rethinking package, models from
the rethinking package are also supported. Note that in
tidybayes.rethinking, tidy_link takes the place of
add_fitted_draws() and tidy_sim takes the place of
add_predicted_draws().
To include basic support for new models, one need only implement the tidy_draws() generic function
for that model.
To include support for estimation and prediction, one must implement the fitted_draws() and
predicted_draws() generic functions.