Summarize grid search trials

module_trials() summarizes the trials that optimize_grid() recorded on a module: how many were run, the best trial, its score and parameters, and the mean and standard error of all trial scores. It reads only the grid-search trials in module$state$trials, which optimize_grid() and GridSearchTeleprompter() write. For other optimizers, use optimization_result() or top_trials().

Usage

module_trials(module, objective = c("maximize", "minimize"))

Arguments

module

A module.

objective

"maximize" (the default) or "minimize": which end of the scores counts as best.

Value

A one-row tibble with columns:

• n_trials: number of trials.

• best_trial: identifier of the best trial.

• best_score: its score.

• mean_score, std_error: mean and standard error of the trial scores.

• best_params: list-column with the best trial's parameters.

• trials: list-column with the full trials tibble.

Without trials, n_trials is 0 and the scores are NA. When every trial failed, a warning is raised and the scores are NA.

See also

Other grid search: GridSearchTeleprompter(), module_metrics(), module_parameters(), optimize_grid()

Examples

mod <- module(signature("text -> sentiment"))
module_trials(mod)
#> # A tibble: 1 × 7
#>   n_trials best_trial best_score mean_score std_error best_params trials  
#>      <int>      <int>      <dbl>      <dbl>     <dbl> <list>      <list>  
#> 1        0         NA         NA         NA        NA <NULL>      <tibble>

if (FALSE) { # \dontrun{
optimize_grid(
  mod,
  data = devset,
  metric = metric_exact_match(field = "sentiment"),
  grid = data.frame(reasoning_effort = c("none", "low", "medium")),
  .llm = ellmer::chat_openai(model = "gpt-6-luna")
)
module_trials(mod)$best_params
} # }