LoginSign Up
← Back to all sparks
D

discretefdr

ANALYTICS
Velocity0.0

R package discretefdr by disohda — release notes from GitHub.

github.com/disohda/discretefdr ↗

The discrete-data FDR package is being pared into one piece of a larger multiple-testing suite.

multiple-testingfalse-discovery-ratediscrete-statisticsr-packagebiostatisticsmodularization
◆Current state
DiscreteFDR implements false discovery rate procedures adapted for discrete test statistics, where the standard continuous-case corrections are conservative. It now covers a discrete Benjamini-Yekutieli procedure alongside the Benjamini-Hochberg variants it started with, including adaptive versions. Its datasets and test-result classes have been moved out into companion packages, so it increasingly does one job and defers the rest.
◆Where it's heading
The direction is decomposition into a suite. The amnesia dataset went to DiscreteDatasets, summary output now interoperates with the DiscreteTestResults class from DiscreteTests, and match.pvals() stopped being exported — each release trims something that belongs elsewhere. What remains gets methodological additions at a slow, deliberate cadence, with performance work on the step-up procedures that dominate cost when the number of tests is large. Recent activity is maintenance: replacing deprecated calls the package still made of its own siblings. This is a mature statistical package whose release notes are short because the methods underneath them are settled.
◆Prediction
Expect further alignment with the companion packages rather than new procedures, since the last substantive release was already about interoperating with DiscreteTests classes and the most recent one about clearing deprecations.

◆Recent moves

  1. 3mo ago

    Deprecated internal calls replaced

    Housekeeping: functions the package had already deprecated were still being called by fast.Discrete() and the vignettes, and this clears them. No change to results or interfaces.

    View source ↗
  2. 1y ago

    Discrete Benjamini-Yekutieli procedure added

    DBY() extends the package past its Benjamini-Hochberg roots to a procedure valid under arbitrary dependence, the substantive methodological addition in this window. The rest is integration work — named p-values flowing through to summaries, which requires a specific DiscreteTests version — plus a correction to CDF indices after selection that the notes describe as harmless in current usage but latent.

    View source ↗
  3. 1y ago

    Datasets split out and step-up procedures sped up

    The clearest instance of the suite strategy: the amnesia dataset moves to DiscreteDatasets and match.pvals() is withdrawn from the public interface. A histogram mode parameter handles conditional p-value selection, and step-up procedures get faster where it matters most, at large numbers of tests.

    View source ↗