uniformNoiseGMM(): Gaussian mixture with a uniform noise component (Banfield & Raftery style), validated against mclust’s noise-component model. Each observation’s posterior probability of the noise component can serve as an outlier score.
All fitters now return the last iterate with converged = FALSE and a warning when maxiter is reached, instead of silently returning NULL. Results also carry an iter count.
testthat suite covering all three fitters, including the beta -> 0 reduction of robustGMM() to the MLE and its resistance to a gross outlier.
Completed documentation for loo_cvm_div() and the fitters; pkgdown reference index organized by topic.
Bug fixes
rnormix(method = "fix") returned more than n samples for k >= 3 components.
robustGMM 0.1.0
Initial version: robustGMM() implementing the density power (beta-)divergence EM estimator of Fujisawa & Eguchi (2006), loo_cvm_div() for tuning-parameter selection, and normal mixture utilities (dnormix(), pnormix(), rnormix(), lse()).