Changelog • rebalancedNoise

rebalancedNoise 0.3.1

Batch tabulation in $perturb() and rn_perturb(): when multiple variables are passed, the table is now aggregated only once via a single makeProblem()/sdcProb2df() call instead of once per variable; this eliminates redundant hierarchy computations and significantly speeds up perturbation of large tables with many variables
Cell sensitivity with n_threshold-only rules is computed once per table instead of once per variable, and per-variable sensitivity for dominance rules no longer requires repeated full-table merges
$perturb() now derives its skip cache from the stored result tables, so already calculated variables are also detected for tables restored via rn_setup() from an export
Fix: force = TRUE in $perturb() no longer corrupts an existing result table with duplicated .x/.y columns; stale columns of the re-computed variables are dropped before merging
Fix: perturbing with dominance rules (p_rule/nk_rule) no longer aborts with a duplicated-strID error coming from .compute_cell_sensitivity()

Functional (pipe) API with typed S3 objects: rn_init() (rn_initialized), rn_rebalance() (rn_rebalanced), and rn_perturb() (rn_perturbed). Each step strictly validates its input class and returns an object consumable by the next step
rn_format(): Retrieve a rn_perturbed result table in wide or long format (functional counterpart of $get_results())
rn_summarize(): Summarize a rn_perturbed result table (functional counterpart of $summarize())
print() and as.data.table() methods for all three pipeline classes
New R6 bridge methods get_state() (returns rn_rebalanced/rn_initialized) and get_table(name) (returns rn_perturbed) to feed R6 engine state into the functional API and vice versa
rn_setup() and the R6 $initialize() now reuse the exported rn_init() internally
New R6 method $reset(): returns the object to its initialized state (strips direction_rebalanced, strID and is_sens_* columns, restores original row order, clears cached results; optionally replaces sensitive_params)

$rebalance() (R6) can only be performed once per object; calling it again aborts with a hint to call $reset() first. The functional rn_rebalance() remains stateless and can be applied repeatedly

Calling rn_rebalance() a second time on already-rebalanced data no longer fails with “object ‘strID’ not found”; a stale strID column from a previous run is dropped before the structural merge
Importing an in-memory rebalancedNoise_ExportData object now applies the exported sensitive_params; previously the override silently failed for object imports (file path imports were unaffected)
Wide-format result tables now place the is_sens_<variable> column next to its variable’s value columns instead of at the end

Removed unused progress dependency and replaced the single glue usage with base R; $export() and $list_tables() messages now respect SDC_LOG_LEVEL = "OFF"
Cell rebalancing now runs as a C++ kernel (rebalance_cells_cpp()) with OpenMP parallelization over base cells, replacing a per-cell R loop; rebalancing on 1M records is about 7 times faster and output is identical (deterministic, no RNG involved)
R6 engine and functional API share the same core: perturbation/tabulation (.perturb_tabulate()), variable merging (.merge_var_into_table()), formatting (.process_result_table()), and summaries (.summarize_entry())
Removed deprecated internal .compute_sensitivity(); single package-wide definition of %||%

Vignettes split into two: “Getting Started with rebalancedNoise” (overview of the method and both interfaces with parallel R6 and functional examples) and “Perturbation Workflows” (detailed features shown in both styles: multiple tables, linear combinations, naming and caching, rounding, evaluation, export and import, parallelization)

New round argument in $perturb() (default FALSE): rounds perturbed microdata values with round() before tabulation so all published cell values are whole numbers
$export() method: Save rebalanced state and results to RDS file or export object
rn_setup() auto-detection: Import from export objects or file paths
$get_microdata(): Extract microdata with direction columns and noise multipliers
$list_tables(): List all perturbed tables with dimensions and variables

direction_rebalanced is now guaranteed to be strictly integer (-1/+1): directions are taken from the balancing algorithm instead of being back-solved from perturbed values, which previously produced NaN for zero values or zero noise multipliers and floating-point drift
$rebalance() aborts with a clear error if any rebalanced direction is invalid (NA/NaN/Inf/non-integer)
Import validation now also checks direction_rebalanced values in export objects

Initial release of the rebalancedNoise package.
Implementation of the EZS method (Sabolova et al., 2025) for magnitude tables.
Contains the dynamic rebalancing algorithm to minimize noise in non-sensitive cells and preserve additive consistency across hierarchies.
Added rebalancedNoise R6 class for integrated SDC workflows.
- Method $perturb(): Perturbs a numeric variable and performs rebalancing
- Method get_results(): Allows retrieval of results
  - Supports both Wide and Long output formats
  - Standardized column naming convention (e.g., [var]_orig, [var]_pert).
  - Includes metadata variables: is_internal to identify internal and is_sens to track sensitivity status.
  - Computes percentage deviations before and after the rebalancing step.
Added rn_setup() to initialize a rebalancedNoise object with support for multiple numerical variables.
Added a package vignette to get started vignette("getting-started", package = "rebalancedNoise")
Performance: Integrated OpenMP support to accelerate the rebalancing procedure written in C++.
- can be configured using argument n_threads in rn_setup() or options(rn_threads) / Sys.getenv("rn_threads")