enw_pathfinder: Fit a CmdStan model using the pathfinder algorithm

In seabbs/epinowcast: Flexible Hierarchical Nowcasting

Fit a CmdStan model using the pathfinder algorithm

Description

For more information on the pathfinder algorithm see the CmdStan documentation. # nolint

Usage

enw_pathfinder(
  data,
  model = epinowcast::enw_model(),
  diagnostics = TRUE,
  init = NULL,
  ...
)

Arguments

data	A list of data as produced by model modules (for example enw_expectation(), enw_obs(), etc.) and as required for use the model being used.
model	A cmdstanr model object as loaded by enw_model() or as supplied by the user.
diagnostics	Logical, defaults to TRUE. Should fitting diagnostics be returned as a data.frame.
init	A list of initial values or a function to generate initial values. If not provided, the model will attempt to generate initial values
...	Additional parameters to be passed to CmdStanModel$pathfinder().

Details

Note that the threads_per_chain argument is renamed to num_threads to match the CmdStanModel$pathfinder() method.

This fitting method is faster but more approximate than the NUTS sampler used in enw_sample() and as such is recommended for use in exploratory analysis and model development.

Value

A data.table containing the fit, data, and fit_args. If diagnostics is TRUE, it also includes the run_time column with the timing information.

See Also

Functions used to help convert models into the format required for stan enw_formula_as_data_list(), enw_get_cache(), enw_model(), enw_priors_as_data_list(), enw_replace_priors(), enw_sample(), enw_set_cache(), enw_stan_to_r(), enw_unset_cache(), remove_profiling(), write_stan_files_no_profile()

Examples

pobs <- enw_example("preprocessed")

nowcast <- epinowcast(pobs,
 expectation = enw_expectation(~1, data = pobs),
 fit = enw_fit_opts(enw_pathfinder, pp = TRUE),
 obs = enw_obs(family = "poisson", data = pobs),
)

summary(nowcast)

seabbs/epinowcast documentation built on Feb. 3, 2025, 9:56 p.m.