LoginSign Up
← Back to all sparks
V

vim

ANALYTICS
Velocity0.0

R package vim by statistikat — release notes from GitHub.

github.com/statistikat/vim ↗

Six dormant years end with a correctness audit across VIM's entire imputation surface

r-packagemissing-dataimputationcorrectness-auditstatistical-validity
◆Current state
VIM handles visualization and imputation of missing values in R, with kNN, hot-deck, iterative robust model-based imputation and matching-based methods. Development effectively stopped after 6.0.0 in 2020. Version 7.2.0 arrives in July 2026 as an explicitly framed correctness milestone: MI-properness warnings, ordered-factor preservation, a keep_all_columns option, list returns from irmi(mi>1), repairs to imputeRobust and imputeRobustChain, cellwise IRWLS and initial-weight fixes, and kNN and gowerD mixed-scaling corrections with a weightDist guard.
◆Where it's heading
The release notes describe an audit — Wave 1 plus tail — rather than a feature cycle, and the fixes cluster around statistical validity: whether multiple imputation is proper, whether factor ordering survives, whether distance scaling across mixed variable types is right. Those are the properties users cannot easily verify themselves, so a package correcting them after six years is implicitly restating what its earlier output was worth. The notes also name a forthcoming R Journal paper under the name vimpute, which points at a successor or companion identity.
◆Prediction
The entries call this a stable reference point for a paper and refer to Wave 1, so a further audit wave is the most likely next release; the vimpute naming is worth watching but the entries do not say what it is.

◆Recent moves

  1. 1mo ago

    Correctness audit fixes MI-properness, factor order and distance scaling

    Ends a six-year silence with an audit pass across the imputation surface: MI-properness warnings, ordered-factor preservation, repairs to imputeRobust and the cellwise IRWLS path, and mixed-scaling corrections in kNN and gowerD. Fixes to statistical validity rather than features, which quietly reframes what output from earlier versions was worth.

    View source ↗
  2. 6y ago

    Adds ranger-based imputation, drops survey and GUI support

    Adds rangerImpute() for random-forest imputation while removing survey object support and the VIMGUI exports, and demotes data.table from Depends to Imports. A deliberate narrowing of scope to the imputation core — and the last release before the package went quiet.

    View source ↗
  3. 6y ago

    Adds nine example datasets and splits help pages

    Bundles nine teaching datasets and separates the maxCat() and sampleCat() documentation. Supporting material for the package's instructional role rather than a change to imputation behavior.

    View source ↗
  4. 6y ago

    Adds matchImpute() and random-forest augmented kNN

    Introduces matchImpute() for within-group random imputation, adds a random forest option to kNN(), and fixes distance computation across mixed variable types. That distance bug recurs in the 7.2.0 audit, marking mixed-type distance as a long-running weak point.

    View source ↗
  5. 6y ago

    Ordered factor support and ordinal regression in irmi()

    Adds ordered factor support and switches irmi() to ordered logistic regression for ordinal variables. Ordered-factor handling reappears as a correctness fix in 7.2.0, so this is where a recurring theme in the package first surfaces.

    View source ↗
  6. 6y ago

    Bug fixes for kNN, hotdeck and irmi input handling

    Fixes for data.table interaction in kNN(), labelled variables in hotdeck(), and data.frame input to irmi(), plus a JSS citation. Routine repair work from the package's active period.

    View source ↗