A run store is a directory of fitted runs, one folder per run, grouped by
specification hash. Each folder holds the specification as YAML, the
fitted object and placebo draws, the pool and balance records, and a
one-row summary that refresh_runs_index() collects into an index. A run
reloaded with load_run() can be re-plotted without refitting.
Usage
save_run(run, store)
load_run(path)
refresh_runs_index(store, write = TRUE)
prune_runs(store, keep_latest_per_spec = 3L, dry_run = FALSE)Arguments
- run
An
sdid_runfromfit_spec().- store
Directory of the run store.
- path
A single run's directory (the
pathcolumn of the index).- write
Cache the index as
index.csvin the store.- keep_latest_per_spec
How many runs to keep for each specification.
- dry_run
Report what would be deleted without deleting it.
Value
save_run() returns the run with its path set; load_run() an
sdid_run; refresh_runs_index() a tibble with one row per run;
prune_runs() the rows it deleted (or would delete).
Examples
# \donttest{
store <- tempfile("runs")
run <- fit_spec(sdid_spec(inference = list(replications = 20)), prop99_panel(),
store = store, seed = 1)
idx <- refresh_runs_index(store)
idx[, c("spec_hash", "att", "se")]
#> # A tibble: 1 × 3
#> spec_hash att se
#> <chr> <dbl> <dbl>
#> 1 13393a31bcc0 -15.6 12.3
again <- load_run(idx$path[1])
all.equal(again$fit$att, run$fit$att)
#> [1] TRUE
# }