LoginSign Up
← Back to all sparks
S

sps

INFRA · APIS
Velocity2.5

Sequential Poisson Sampling

github.com/marberts/sps ↗

sps keeps sanding down sequential Poisson sampling rather than adding to it.

r-packagesurvey-samplingsequential-poissonperformanceapi-stability
◆Current state
The package implements sequential Poisson sampling for survey design, covering inclusion probabilities, proportional allocation, replicate weights, and take-all strata. Recent releases are small and tightly scoped: a divisor method helper, an iterator that draws a sample one unit at a time, automatic selection of the replicate-weight parameter, and repeated performance work on inclusion probability calculation.
◆Where it's heading
Development is consolidation rather than expansion. Most releases either speed up an existing routine or remove a decision the user previously had to make by hand, such as picking the smallest parameter that keeps replicate weights non-negative. Documentation and tooling get comparable attention to the algorithms, with a dedicated vignette on inclusion probabilities and a recent switch of test and documentation infrastructure. The API surface has been essentially stable across the window.
◆Prediction
Expect continued small ergonomic and performance releases against the existing function set rather than new sampling designs, which is the pattern every release in this window follows.

◆Recent moves

  1. 29d ago

    Documentation polish; switches to tinytest and litedown

    Documentation improvements and a switch to tinytest and litedown for testing and rendering. Maintenance infrastructure with no effect on sampling behaviour.

    View source ↗
  2. 9mo ago

    Fixes extra argument handling in sps_iterator()

    A one-line fix so extra arguments pass through the sampling iterator correctly. Cleanup on the iterator introduced two releases earlier.

    View source ↗
  3. 11mo ago

    Adds divisor_method() and a one-unit-at-a-time sampling iterator

    Adds a helper for generating divisor functions by name and an iterator that draws a sample one unit at a time, alongside a faster allocation routine. The iterator is the most substantive addition in the window, since it allows sampling to be driven incrementally rather than in one call.

    View source ↗
  4. 1y ago

    Adds an inclusion-probability vignette and faster partial sorting

    A vignette explaining how inclusion probabilities are calculated, plus partial sorting through kit::topn() when available. Documentation and a contained optimisation.

    View source ↗
  5. 1y ago

    Automatic tau selection for replicate weights

    Replicate weight generation now picks the smallest parameter that avoids negative weights by default, removing a manual tuning step. Single-stratum inclusion probability calculation is also faster, which benefits the main sampling entry points.

    View source ↗
  6. 2y ago

    Adds becomes_ta() for take-all stratum sample sizes

    Adds a function to determine the sample size at which a unit enters the take-all stratum, filling a gap in the allocation workflow. Remaining changes are internal class instantiation with no user-visible effect.

    View source ↗