Summarize an optimization result

optimization_summary() condenses optimization_result() into the numbers most often reported: the number of trials, the best score and parameters, the score range, the total cost and the improvement over the baseline.

Usage

optimization_summary(module)

# S3 method for class 'dsprrr_optimization_summary'
print(x, ...)

Arguments

module

A module returned by compile() or modified by optimize_grid().

x

A dsprrr_optimization_summary object.

...

Unused.

Value

A dsprrr_optimization_summary list with n_trials, best_score, best_trial, best_params, score_range (minimum and maximum trial scores), total_cost (when trials record it), improvement (best score minus the baseline or first score) and compiled. print() shows it and returns it invisibly.

See also

Other optimization results: apply_best_config(), best_demos(), best_params(), config_diff(), export_module_code(), optimization_result(), top_trials()

Examples

classifier <- module(signature("text -> sentiment"))
trainset <- data.frame(
  text = c("I love it!", "Terrible experience", "It's okay"),
  sentiment = c("positive", "negative", "neutral")
)
compiled <- compile(classifier, LabeledFewShot(k = 2L), trainset)
optimization_summary(compiled)$best_params
#> $k
#> [1] 2
#> 

if (FALSE) { # \dontrun{
# After a search with several trials, printing gives an overview
optimize_grid(
  classifier,
  data = trainset,
  metric = metric_exact_match(field = "sentiment"),
  grid = data.frame(reasoning_effort = c("none", "low")),
  .llm = ellmer::chat_openai(model = "gpt-6-luna")
)
optimization_summary(classifier)
} # }