colvR 0.2.2
Changes
- Removed internal test function cpp_test_packing() which triggered a
-Wmismatched-new-delete warning with GCC 16.
colvR 0.2.1
Changes
- Explanation of acronyms
- Removal of console output from Miss.ZIPLNPCA_Steps()
colvR 0.2.0
Breaking changes
Predictions.marginales() has been removed: use
Predictions(type = "marginal").select_model_bic() now returns a list with the
criterion values, the selected latent dimension (q_star)
and the corresponding fit (fit_star), instead of a vector
of BIC values.- In
Miss.PLNPCA(), q is now the third
argument, as in Miss.ZIPLNPCA().
- The example dataset
fuligule_milouin has been replaced
by waterbird, with anonymized sites and years and without
the latitude covariate.
New features
Predictions() gains the arguments type
(conditional or marginal intervals), probs (quantile
levels), output (draws, latent variables and parameter
draws in addition to the intervals), cl (parallel VE steps)
and verbose.- New function
ICL(), the Integrated Completed
Likelihood.
select_model_bic() gains the arguments
criterion (BIC or ICL) and model.BIC() gains the argument model
("ZIPLNPCA" or "PLNPCA").
Bug fixes
BIC() now counts 2d regression parameters
for the ZIP-PLN-PCA model, and d or pd for the
PLN-PCA model, depending on the design.select_model_bic() now works with non-contiguous values
of latents.- In the
Miss.ZIPLNPCA*() functions,
pred$predicted is now the ZIP expectation
xi * A, consistent with imputed.
Predictions() no longer fails when
q = 1.
Other changes
- All examples are now runnable, on small simulated data.
- New unit tests for the selection criteria,
select_model_bic(), Predictions() and
Miss.ZIPLNPCA().
- All messages are now in English.