run_mme_benchmark: Run Experimental MME Benchmark Suite

In sommer: Solving Mixed Model Equations in R

Run Experimental MME Benchmark Suite

Description

Executes a battery of workloads across candidate solver engines, measuring wall-clock time, iterations, peak memory (if available), active numerical paths, and asserting numerical parity.

Usage

run_mme_benchmark(
  workloads,
  engines = NULL,
  replications = 3L,
  verifyParity = TRUE,
  tolerance = 1e-05,
  nIters = 5L,
  verbose = TRUE
)

Arguments

workloads	A list of mme_workload objects (or a single workload).
engines	A named list of mmes_control objects. Defaults to comparing baseline (standard auto solver) against candidate experimental toggles.
replications	Number of timed repetitions (warmup run excluded). Default is 3.
verifyParity	Logical; whether to assert that experimental candidates match the reference engine in final likelihood, variance components, and BLUPs. Default is TRUE.
tolerance	Numerical tolerance for parity verification. Default is 1e-5.
nIters	Maximum REML/ML iterations for benchmark fits. Default is 5.
verbose	Logical; whether to print progress. Default is TRUE.

Value

An object of class "sommer_mme_benchmark". Warmup and timed-run failures retain the same result columns with an error message. The converged column records the fitted model's convergence flag. parity_ok is NA when parity is not checked or the reference engine failed; numerical comparisons use final likelihoods and estimates, not the lengths of their optimization histories.

sommer documentation built on Oct. 24, 2026, 1:08 a.m.