Fit a reproducible ensemble of self-organizing maps — fit_som

Every model receives preprocessing parameters estimated only from its own analysis rows. Failures are recorded instead of silently discarded. Topographic error is computed during fitting so duplicate full processed matrices need not be retained in every ensemble member.

Usage

fit_som_ensemble(
  data,
  spec,
  resamples = NULL,
  preprocess = som_preprocess(),
  keep_models = TRUE,
  fail_fast = FALSE,
  parallel = FALSE
)

Arguments

data: A som_data object.
spec: A som_spec object.
resamples: A som_resamples object. The default fits the full data.
preprocess: One som_preprocess object or a named object per layer.
keep_models: Whether to retain fitted kohonen objects.
fail_fast: Whether to stop at the first model failure.
parallel: Whether to distribute ensemble members through the current future plan using future.apply. The package never changes that plan. Explicit model seeds retain reproducibility across sequential and future execution. Avoid combining worker-level parallelism with spec$cores > 1 unless nested parallelism is intended.

Value

A som_ensemble object.

Examples

data <- simulate_som_scenario("clusters", n = 45, p = 3, seed = 1)
specification <- som_spec(c(3, 2), seeds = 1:2, rlen = 10, k = 2)
ensemble <- fit_som_ensemble(data, specification, keep_models = FALSE)
ensemble
#> <som_ensemble>
#>   attempted: 2 
#>   succeeded: 2 
#>   failed   : 0 
#>   warnings : 0 
#>   samples  : 45