Skip to content

exametrika v1.10.0

Choose a tag to compare

View all tags
@kosugitti kosugitti released this 19 Mar 06:45
· 138 commits to main since this release
v1.10.0
bac336d

Highlights

  • Nominal IRM (Biclustering_IRM.nominal): Dirichlet-Multinomial collapsed Gibbs sampler for nominal/polytomous data
  • Ordinal IRM (Biclustering_IRM.ordinal): Shared Gibbs core + ordinal EM with cumulative normalization
  • Biclustering_IRM() S3 generic: Auto-dispatches to .binary, .nominal, .ordinal based on data type
  • Confirmatory LCA/LRA: conf parameter for test equating with label-based item matching
  • CAIC formula correction: Fixed to match Bozdogan (1987) original definition
  • Test suite modernization: Migrated from Excel/tidyverse to CSV/base-R (873 tests, 85 reference CSVs)

Bug Fixes

  • CAIC penalty term: log(n+1) → log(n)+1 (Bozdogan 1987)
  • GridSearch index parameter: alias support, early validation, log-likelihood direction
  • LRA.ordinal/rated category computation (apply() matrix-vs-list)
  • GRM ItemFitIndices (apply()/table() same class of bug)
  • dataFormat robustness: CA auto-detection, ID handling, missing values, drop=FALSE
  • BINET g_list/adj_list input path and length validation
  • Biclustering_IRM log-probability normalization (numerical stability)
  • Biclustering_IRM t(apply()) dimension drop when ncls=1 or nfld=1
  • Biclustering.nominal stale log-likelihood in model fit
  • J20S400 dataset: response.type nominal → binary
  • Various other fixes (see NEWS.md for full details)

New Features

  • Shared Gibbs core (irm_gibbs_core()): Extracted into R/00_IRM_Gibbs_CORE.R
  • Confirmatory LCA/LRA: conf parameter with label-based item matching, works with GTM and SOM
  • SOM estimation refactoring: Extracted somclus() internal function
  • New plot types: FRP, FCRP, FCBR, ScoreField, RRV for polytomous Biclustering
  • FRPIndex: Field Reference Profile indices for ordinal Biclustering
  • Sample datasets: J35S500 (ordinal), J20S600 (nominal)
  • Return value structure unification: snake_case fields, 16-field ModelFit structure

Test Suite

  • 26 test files, 321 test blocks, 873 tests — all passing
  • 85 Mathematica reference CSV files for cross-validation
  • Zero external test dependencies beyond testthat
  • R CMD check: 0 errors | 0 warnings | 0 notes

See NEWS.md for full changelog.

Assets 2
Loading