LoginSign Up
← Back to all sparks
S

sdsfun

ANALYTICS
Velocity0.0

R package sdsfun by stscl — release notes from GitHub.

github.com/stscl/sdsfun ↗

A spatial-statistics utility package exists to be depended on, and is built accordingly.

spatial-statisticsgeodetectorspatial-clusteringrcppshared-infrastructurer-package
◆Current state
sdsfun collects spatial data science utilities — neighbour lists, spatial constrained clustering, discretization, dummy variable generation, geographical detector statistics and projection helpers — with the computationally heavy parts implemented in Rcpp. It was assembled quickly across late 2024, adding a function set roughly every three weeks, and has slowed since to a couple of releases a year. The most recent work is corrective: no longer initializing the RNG state at load, fixing matrix inputs misread as vectors, and clearing an Armadillo deprecation.
◆Where it's heading
This is infrastructure for a family of packages rather than an end-user tool, and the changelog says so directly — functions were added to support gdverse and sesp, and moran_test was migrated in from geocomplexity. That migration pattern is the defining move: capability consolidates here so the downstream packages can share it instead of each carrying its own copy. Growth has slowed as that consolidation completed, leaving correctness and dependency upkeep.
◆Prediction
Given the package moves when its dependents need something, the next release most likely brings in another shared function or responds to a downstream requirement rather than following its own plan. Armadillo and CRAN check changes remain the reliable source of maintenance work.

◆Recent moves

  1. 10mo ago

    Package load stops touching the RNG state

    Small in size, real in consequence: a package that seeded the RNG merely by being loaded could shift results in any downstream analysis that set a seed and then attached it. Alongside it, cor_test no longer misreads matrix inputs as vectors, and an Armadillo deprecation is cleared.

    View source ↗
  2. 1y ago

    Unified partial correlation testing and head/tails discretization

    cor_test consolidates correlation and partial correlation testing behind one interface, continuing the pattern of centralising what dependents would otherwise duplicate. The head/tails breaking method adds a discretization scheme suited to heavy-tailed spatial distributions, where equal-interval and quantile breaks perform poorly.

    View source ↗
  3. 1y ago

    Missing-value handling added to linear trend removal

    A single targeted fix to how rm_lineartrend() treats NA values. No other change in the release.

    View source ↗
  4. 1y ago

    Covariate-based detrending and long-to-matrix spatial reshaping

    Extends trend removal from a simple linear fit to one conditioned on a covariate, and adds conversion of long-form tables into matrices matching two-dimensional spatial positions. Both are preparation steps that every dependent package would otherwise implement separately.

    View source ↗
  5. 1y ago

    Spatially constrained hierarchical clustering and SPADE estimation

    hclustgeo_disc brings clustering with spatial soft constraints into the shared toolkit, and spade_psd adds a fast SPADE model estimator to sit beside the geodetector routine added a week earlier. Part of the dense late-2024 build-out where the package acquired most of its surface.

    View source ↗
  6. 1y ago

    Fast geodetector q-value estimator added

    A single-function release adding fast estimation of the geodetector q-value, the central statistic of the geographical detector method. Its arrival as a standalone routine here rather than inside gdverse is the consolidation pattern in miniature.

    View source ↗