Skip to contents

Summarise benchmark simulation results into win/tie/loss tables

Source: R/benchmark_comparators.R
summarise_benchmark_results.Rd

summarise_benchmark_results() turns rows from run_coverage_simulations() or benchmark_standard_models()-style outputs into compact evidence tables. Each metric is scored according to its estimand: lower error is better, coverage targets 0.95, calibration/tail errors target 0, and runtime is lower-is-better.

Usage

summarise_benchmark_results(
  results,
  metrics = NULL,
  group_cols = c("family", "copula", "design", "n_subject", "n_time", "dependence",
    "missingness", "start_mode"),
  tie_tolerance = 0.05,
  absolute_tolerance = 1e-08
)

Arguments

results

Data frame returned by run_coverage_simulations().

metrics

Optional metric columns to summarise. Defaults to all known benchmark metrics present in results.

group_cols

Columns defining one simulation case before method-level comparison.

tie_tolerance

Relative gap from the case-best score treated as a tie.

absolute_tolerance

Absolute score gap treated as a tie/win.

Value

An object of class gamlss_longitudinal_benchmark_summary with summary, case_results, and metric_catalog components.