Execute a single simulation task

Runs one task: generate data, fit model, compute metrics. All errors are captured and converted to task results.

Usage

run_task(
  task,
  config_spec,
  fitter,
  metrics,
  retain = c("metrics", "diagnostics")
)

Arguments

task

A task specification list (from get_task_spec) containing:

• task_id: Unique task identifier string

• data_spec: Data generation specification

• fit_spec: Model fitting specification

• rng_seed: Precomputed RNG seed for this task

• task_ctx: Task context with indices (data_idx, fit_idx, rep_idx)

config_spec

Plain list config spec (for worker transport) containing:

• data_generator: Function to generate data

• Other configuration parameters

fitter

S7 Fitter object that implements the fit(), predict_fit(), log_lik(), and loo() methods

metrics

List of S7 Metric objects to compute

retain

Character vector of what to retain in results. Options:

• "metrics": Always retained (default)

• "diagnostics": Fit diagnostics (default)

• "fit": The raw fit object

• "draws": Posterior draws matrix

Value

A bayesim_task_result S3 object containing:

• task_id: Task identifier

• status: "success" or "failed"

• metrics: Named list of computed metrics (if successful)

• diagnostics: Named list of fit diagnostics (if successful)

• timing: List with total elapsed time

• warnings: Character vector of warning messages

• error: Error information (if failed)

Details

The task execution follows these steps:

1. Set the RNG state for deterministic reproducibility

2. Generate data using the data_generator function

3. Validate the data bundle

4. Fit the model using the fitter

5. Build context for metric computation (predictions, log_lik, loo)

6. Compute all metrics

7. Apply retention policy to manage memory

Error handling:

• Data generation errors are normalized to bayesim_data_error

• Fit errors are normalized to bayesim_fit_error

• Metric errors are handled in compute_all_metrics()

• Fatal errors (config, contract) propagate and stop the simulation

Examples

if (FALSE) { # \dontrun{
# Create a task specification
task <- list(
  task_id = "d001_f001_r00001",
  data_spec = list(n = 100, effect_size = 0.5),
  fit_spec = list(prior = "normal(0, 1)"),
  rng_seed = create_task_rng_streams(42, 1)[[1]],
  task_ctx = list(data_idx = 1, fit_idx = 1, rep_idx = 1)
)

# Run the task
result <- run_task(
  task = task,
  config_spec = config_spec,
  fitter = my_fitter,
  metrics = list(rmse_metric, coverage_metric),
  retain = c("metrics", "diagnostics")
)

# Check result
if (result$status == "success") {
  print(result$metrics)
} else {
print(result$error$error_message)
}
} # }