Core optimizer: assign samples to groups to balance mean + SD (Gini loss) — balance_partition_core • Rtoolset

Internal/core function that performs the random search and returns the best grouping. No plotting and no file writing.

The objective minimized is: $$loss = Gini(group\_means) + lambda * Gini(group\_sds)$$

Usage

balance_partition_core(
  score,
  group_sizes,
  lambda = 1,
  B = 50000,
  method = c("blocked_permute", "random_assign"),
  seed = NULL
)

Arguments

score: Numeric vector (length n). The scalar to balance.
group_sizes: Integer vector of target group sizes. Must sum to length(score).
lambda: Numeric >= 0. Weight on SD-balance term. Default is 1.
B: Integer >= 1. Number of candidate partitions to evaluate.
method: Character. One of "blocked_permute" or "random_assign".
seed: Optional integer seed.

Value

A list with:

groups: integer vector (length n) of group labels 1..K
loss: best achieved loss
group_sizes: integer vector of target sizes used