Simulate Custom 'simulate_custom' is the interface for pmsims at the most basic level. It performs no processing of arguments and allows all possible options to be customised.

Usage

simulate_custom(
  data_function = NULL,
  model_function = NULL,
  metric_function = NULL,
  target_performance,
  c_statistic,
  mean_or_assurance = "assurance",
  test_n = 30000,
  min_sample_size,
  max_sample_size,
  n_reps_total = NULL,
  n_reps_per = 50,
  se_final = NULL,
  n_init = 4,
  method = "mlpwr",
  verbose = FALSE,
  ...
)

Arguments

data_function: A function that returns datasets. Must have a single argument, n, which controls the sample size.
model_function: A function that fits models to the data. Takes the data object returned by data_function as its only argument.
metric_function: A function that returns a performance metric. Must take test data, a fitted model and a model function as arguments. Must return a single value.
target_performance: Numeric target performance threshold the algorithm must meet or exceed.
c_statistic: Numeric; anticipated large-sample discrimination used when tuning the data generator.
mean_or_assurance: Character string "mean" or "assurance" indicating which criterion defines the minimum sample size.
test_n: Integer size of the test datasets used to evaluate performance (should be large).
min_sample_size: Integer lower bound of the sample-size search region.
max_sample_size: Integer upper bound of the sample-size search region.
n_reps_total: Integer total number of simulation replications allocated to the search.
n_reps_per: Integer number of replications evaluated at each candidate sample size.
se_final: Numeric standard error threshold used for early stopping (supply either n_reps_total or se_final).
n_init: Integer number of initial sample sizes explored before the main search algorithm begins.
method: Character string selecting the search engine; currently "mlpwr", "bisection", or "mlpwr-bs".
verbose: Logical flag controlling printed progress information.
...: Additional arguments passed to the chosen engine (for example tol for bisection or penalty_weight for GA-based methods).

Value

An object of class "pmsims" containing the estimated minimum sample size and simulation diagnostics.