Skip to contents

News and updates

Gao Wang

2026-05-17

Source: vignettes/announcements.Rmd
announcements.Rmd

Version 0.2.0

S3 implementation

mvsusieR has been re-implemented using R’s S3 object system, replacing the previous R6-based architecture (in production during 2019–2025). The S3 implementation provides the same functionality with a cleaner, more maintainable codebase with core algorithmatic implementation shared with susieR, making it a required dependency for this package.

Rcpp/C++ acceleration

Performance-critical operations are implemented in C++ via Rcpp/RcppArmadillo to provide substantial speedups.

Changed defaults

Users upgrading from 0.1.x (R6) should note the following changes to default parameter values:

  • estimate_residual_variance now defaults to TRUE in mvsusie() (was FALSE). The residual variance is estimated from the data unless explicitly set. The default remains FALSE in mvsusie_ss().
  • estimate_prior_method now defaults to "optim" (was "EM"). The optimization-based method is now faster and more accurate for mixture priors.
  • estimate_prior_mixture_weights now defaults to TRUE. Mixture prior component weights are re-estimated during fitting via the "mixsqp" algorithm by default (can also be "EM").
  • precompute_cache now defaults to TRUE (was FALSE; previously named precompute_covariances). Caches eigendecompositions and other intermediate quantities to accelerate the IBSS fitting algorithm. If memory is a concern with very large numbers of mixture components, set to FALSE at the cost of slower computation.
  • missing_y_method with default to "approximate", replacing the previous approximate boolean parameter (also default to TRUE).
  • verbose replaces the numeric verbosity parameter and defaults to TRUE (logical).
  • model_init replaces s_init for model initialization (consistent with current susieR).

References