Skip to contents

Run the stress-test grid over the benchmark

Source: R/05-perturb.R
fx_stress_grid.Rd

Perturbs every column in data along each requested dimension and level, then re-scores with the benchmark's per-case scorer. A single shared clean baseline per case is emitted (dimension = "(clean)"), so decay is measured relative to it.

Usage

fx_stress_grid(
  data = fx_benchmark_data(),
  dimensions = NULL,
  router = "auto",
  reps = 1L,
  seed = 1L,
  max_n = 300L,
  contamination_sweep = FALSE,
  length_min_unique = 10L,
  progress = FALSE
)

Arguments

data

A benchmark data frame from fx_benchmark_data().

dimensions

Which dimensions to run (default: all).

router

Router passed to guess_data_type().

reps

Repetitions per (case, dimension, level) for stochastic dimensions.

seed

Base seed (per-cell seeds are derived from it, so runs reproduce).

max_n

Cap on the reconstructed / resampled column length.

contamination_sweep

If TRUE, also run contamination at 0.4 and 0.5 to locate the breaking point (see fx_contamination_curve()).

length_min_unique

Minimum distinct values for the length dimension to apply to a case (others are recorded with applicable = FALSE).

progress

If TRUE, print a progress dot every 200 runs.

Value

A long data frame: case_id, gold_data_type, gold_variant_id, dimension, level, rep, applicable, foil_type, the per-level *_correct flags, headline, outcome, and elapsed_sec.

See also

fx_stress_summary(), fx_contamination_curve(), fx_perturb()