Aggregate simulation results per condition

Computes per-condition aggregates of the wide summary tibble: mean and median of each metric column, Monte Carlo standard errors (MCSE), replicate counts, and failure/convergence-failure rates. Returns a tidy tibble with one row per condition.

Aggregation follows each metric's declared summary_type (E4; see Metric): "mean" columns get a sd / sqrt(n) MCSE, "proportion" columns (e.g. coverage) get sqrt(p(1-p) / n), and "none" columns (e.g. SBC ranks) are excluded from aggregation. Columns from unknown or user-defined sources default to "mean". MCSE formulas follow rsimsum (Gasparini, 2018).

Wide summaries: several metrics legitimately flatten to dozens of columns each, so the default aggregation can return 100+ columns. Nothing is ever dropped or truncated. Narrow the output with the metrics argument, and discover a single metric's flattened columns with metric_cols(). In interactive sessions only, a wide default call prints a one-line hint pointing at these; programmatic and noninteractive use is always silent.

Usage

summarize_simulation(result, by = NULL, metrics = NULL)

Arguments

result

A bayesim_simulation_result object (uses result$summary), or a data.frame/tibble of per-task metrics. Passing the full result is preferred: it carries the metrics' summary_type declarations.

by

Character vector of grouping columns (conditions). Defaults to the data_*/fit_* grid columns found in the summary plus any other non-numeric condition columns (excluding task_id and status).

metrics

Character vector of metric columns to aggregate. Defaults to all numeric columns not in by and not metadata (task_id, rep_idx, status, *timing*).

Value

A tibble with one row per condition: the by columns, then for each metric <m>_n_used, <m>_mean, <m>_median, <m>_sd, <m>_mcse, plus n_reps, n_failed, failure_rate. <m>_n_used is the number of finite values used for that metric; failed or non-finite metric values do not contribute to its aggregate.

Examples

if (FALSE) { # \dontrun{
result <- run_simulation(config, progress = FALSE)
summarize_simulation(result)
summarize_simulation(result, by = "model", metrics = "rmse__value")
} # }