NEWS.md

In bivarhr: Bivariate Hurdle Regression with Bayesian Model Averaging

bivarhr 0.1.6

CRAN compliance

• All packages listed under Suggests are now used conditionally, as required by "Writing R Extensions". Every function that relies on a suggested package (bnlearn, sensemakr, CausalImpact, vars, RTransferEntropy, MASS, openxlsx) now checks for it with requireNamespace() and raises an informative error if it is missing.
• All \donttest examples that exercise a suggested package are now wrapped in if (requireNamespace(<pkg>, quietly = TRUE)) (or, for the Stan-based functions, if (interactive() && requireNamespace("cmdstanr"))), so they are skipped under _R_CHECK_DEPENDS_ONLY_=true instead of failing. This resolves the CRAN check failure in the run_hmm example.
• run_hmm() no longer depends on depmixS4 (which was archived from CRAN on 2026-07-04). The bivariate Poisson HMM is now fitted by a self-contained base-R implementation of the Baum-Welch (EM) algorithm with a numerically stable log-space forward-backward pass, multi-start restarts, and Viterbi decoding. depmixS4 has been dropped from Suggests.

Robustness improvements

• Output directories are now explicit function arguments (dir_csv, dir_out, dir_figs) instead of global variables read from the caller's environment. They default to NULL, in which case the functions no longer write to disk. Affected functions: run_hmm(), run_dbn(), run_varx(), run_sensemakr(), run_synth_bsts(), run_transfer_entropy(), run_eba(), rolling_oos().
• run_eba() now takes control_combos as an explicit argument instead of reading it from the global environment.
• run_hmm() now calls depmixS4::fit() explicitly, so it works without attaching depmixS4 to the search path.
• run_varx() now coerces its input to a data.table internally, so it accepts a plain data.frame.
• Fixed a latent bug in smoketest_floor_elpd_invariance() that called a non-existent fit_one_internal(); it now calls fit_one().
• Removed a duplicate definition of export_results_xlsx() that shadowed the fully featured implementation.
• Dropped unused ggplot2 and bayesplot entries from Suggests.

bivarhr 0.1.0

• Initial package release.
• Bivariate hurdle model implementation.
• Time-series analysis functions.

bivarhr documentation built on July 7, 2026, 1:06 a.m.