R/stan-summary-log.R
stan_summary_log.RdThere are two entry points for generating the summary log:
stan_summary_log(): return a tibble the summarizes the Stan models under
the specified directory
stan_add_summary(): join summary columns to a bbr::run_log() tibble.
At least one row of the run log tibble must have a Stan model.
stan_summary_log(
.base_dir,
.recurse = FALSE,
.include = NULL,
variables = "lp__",
summary_fns = NULL
)
stan_add_summary(.log_df, variables = "lp__", summary_fns = NULL)Base directory to look in for models.
If FALSE, the default, only include models in .base_dir
but not subdirectories. This is passed through to fs::dir_ls() –
If TRUE recurse fully, if a positive number the number of levels to recurse.
A character vector specifying which runs or model tags to include in the run log.
Model variables to summarize with summary_fns. This value
is passed through to the CmdStanFit
$summary() method. To summarize these
variables, their draws will be read in. It is valid to specify a variable
that only exists in a subset of the models.
By default this is restricted to "lp__", but you will likely want to extend
this with a few variables of interest. $summary() will use all
variables if you pass NULL, but it's recommended to keep the set small
for performance reasons and because each variable adds a number of columns
equal to the number of summary_fns (or possibly more, if a summary
function returns a named vector). To disable extracting and summarizing the
draws entirely, set summary_fns to an empty list.
A list of summary or diagnostic functions passed to
posterior::summarize_draws(). The result will be included as a new
column, {variable}_{name}, where "name" is determined by the names of the
returned vector or otherwise by the name of the summary_fns item.
With the default value of NULL, the following functions are used:
mean(), stats::median(), posterior::quantile2(), posterior::rhat(),
posterior::ess_bulk(), and posterior::ess_tail(). Pass an empty list to
disable all variable summaries.
a bbi_run_log_df tibble (the output of run_log())
A tibble. For stan_summary_log(), the tibble will have a row for
each Stan model collected from under the directory. For
stan_add_summary(), the tibble will have the same number of rows as
.log_df with additional columns.
The information for each row is primarily extracted from a few methods of the cmdstanr fit object: $metadata, $diagnostic_summary(), and $summary().
https://mc-stan.org/misc/warnings for an overview of diagnostics and warnings