Realize points on a synthetic geometry with explicit random-state control

Coordinates and sampling algorithms retain the existing version-1 draw order. With a seed or explicit state list, the caller's RNG kinds and seed presence/value are restored on success and errors. The opt-in current-stream mode instead advances the caller's stream. No statistical response or dataset identity is created.

Usage

sample.synthetic.geometry(
  geometry,
  sampling,
  n = NULL,
  seed = NULL,
  rng.plan = NULL
)

Arguments

geometry

A supported geometry component.

sampling

A supported sampling component. Legacy G4 is excluded.

n

Positive sample size, or NULL for fixed-size sampling.

seed

Nonnegative integer seed for standalone L'Ecuyer-CMRG sampling. Supply exactly one of seed and rng.plan.

rng.plan

Optional list with exactly version = 1L, order equal to "sampling.frame" or "frame.sampling", and sampling and frame RNG states. States are complete integer .Random.seed vectors for L'Ecuyer-CMRG or Mersenne-Twister with Inversion/Rejection. frame is NULL unless a random frame is requested. State headers retain R's binomial generator field when supported by the running R version; a newer state cannot be replayed on an R version that does not support it. Retry/substream selection belongs to the caller. The standalone seed assigns sampling its initial stream and the random frame the next independent stream. Alternatively, "current" consumes the current R stream directly, sampling before any random frame. This mode never resets RNG kinds or seed and supports the caller's selected generators, including Box-Muller's hidden cached normal value. Draws advance the stream even if a subsequent operation fails. Call set.seed() yourself when reproducibility from a known starting point is needed.

Value

A synthetic_geometry_sample list containing predictors, latent, latent.mask, region, frame.matrix, the specifications, dimensions, intrinsic.dim.by.region and codimension.by.region for simplex strata, declared.regions, observed.regions, n, and sample (the original sampler payload). rng contains the effective plan, states after each draw, and final.state for explicit continuation by a caller. In current-stream mode the plan is "current"; returned state vectors are observations, not complete replay tokens for generators with hidden state. A deterministic draw can leave these vectors NULL if no seed exists. The result has no truth or response fields.

Examples

x <- sample.synthetic.geometry(synthetic.circle(),
  synthetic.sampling.uniform.interval(0, 2 * pi), n = 8, seed = 7)
dim(x$predictors)
#> [1] 8 2