Select flattened metric columns

Selects columns belonging to one metric in bayesim's flattened naming scheme, <metric>__<field> or <metric>__<field>__<param>.

Usage

metric_cols(x, metric, fields = NULL, as = c("names", "long"))

Arguments

x

A bayesim simulation result (with a summary component), or a per-task summary data frame.

metric

Required character scalar naming the metric prefix.

fields

Optional character vector restricting the field segment.

as

Output form: "names" returns a named character vector suitable for dplyr::all_of(); "long" returns one row per task and metric value.

Value

For as = "names", a named character vector whose values are full column names and whose names are suffixes following the metric prefix. For as = "long", a tibble with optional task_id, field, param, and value columns.

Examples

fitter <- LinearRegressionFitter()
summary <- data.frame(
  task_id = "task_1",
  posterior_summary__mean__x = 1.2,
  posterior_summary__sd__x = 0.3,
  check.names = FALSE
)
metric_cols(summary, "posterior_summary")
#>                      mean__x                        sd__x 
#> "posterior_summary__mean__x"   "posterior_summary__sd__x" 
metric_cols(summary, "posterior_summary", fields = "mean", as = "long")
#> # A tibble: 1 × 4
#>   task_id field param value
#>   <chr>   <chr> <chr> <dbl>
#> 1 task_1  mean  x       1.2