← Back to all sparks
G

gdverse

ANALYTICS
Velocity0.0

R package gdverse by stscl — release notes from GitHub.

gdverse is turning geographical detector methods into inference, not just point estimates.

spatial statisticsgeographical detectorconfidence intervalsreticulater package
Current state
A geographical detector toolkit for spatial stratified heterogeneity, shipping small numbered releases every few months. Recent work centres on statistical rigour: confidence intervals for the q-statistic (experimental in 1.3-2, made more robust in 1.6), reported significance for interaction detection, and a fix for stratification collision in that same interaction path. The rest is Python-interop maintenance — reticulate compatibility, parallel stability in cpd_disc, and dependency configuration.
Where it's heading
The arc is from computing detector statistics to qualifying them. Confidence intervals, significance reporting and non-centrality parameter estimation are all about telling users how much to trust a q-value, which is the gap between a research script and a package other people cite. The Python-dependency work is the recurring tax on that: several releases exist mainly to keep reticulate-backed models passing checks.
Prediction
Expect the experimental q-statistic confidence intervals to be promoted to a stable, documented interface across the detector family, since the last two releases have both worked on their robustness and reporting.

Recent moves

  1. 6mo ago

    Interaction detection reports significance; stratification collision fixed

    Interaction detection now reports significance and no longer collides on stratification, and q-statistic confidence intervals get more robust non-centrality parameter estimation. The inference-quality thread continues, alongside a fix that stops loading the package from touching the RNG state.

    View source ↗
  2. 10mo ago

    Python examples wrapped to stop CRAN check failures

    Wraps Python-dependent examples in tryCatch so CRAN checks stop failing. Packaging hygiene for the reticulate dependency, with no change in behaviour.

    View source ↗
  3. 10mo ago

    cpd_disc refactored for parallel stability and reticulate compatibility

    Refactors cpd_disc for parallel stability and reticulate compatibility and updates the Python dependency configuration. Maintenance, but the kind users of the Python-backed models feel directly.

    View source ↗
  4. 1y ago

    Adds package citation metadata

    Adds package citation information. Metadata only.

    View source ↗
  5. 1y ago

    Experimental confidence intervals for the q statistic

    Introduces experimental confidence intervals for the q statistic in factor_detector, renames gd_besttunidisc to gd_optunidisc, and revises pseudo p-value calculation for the SPADE model. The first release where the package starts quantifying uncertainty rather than only reporting q.

    View source ↗
  6. 1y ago

    Plot method bug fixes across four detector models

    Fixes bugs in the plot S3 methods for the gd, opgd, gozh and srsgd models. Rendering fixes with no change to the underlying statistics.

    View source ↗