clustvarsel 2.3.4 (2020-12)

clustvarsel 2.3.3 (2018-11)

Added the final estimated model to the clustvarsel object.
Solved a bug that stop execution in the greedy-backward search when no variables could be removed.

Package version accompanying JSS paper.
Bug fixes in the extreme case no clustering variable is selected using the greedy forward/backward search.

Add optional argument verbose to clustvarsel() for printing steps info during the search.
New print method for clustvarsel objects.
A parallel cluster is automatically stopped unless a registered parallel back end is provided as argument to parallel argument in the clustvarsel() function call.
Add “A quick tour of clustvarsel” vignette.

Version associated with JSS paper submission.
Add explicitly stop of clusters if parallel is used.
Specifically included in the hc() function call the argument name data = ... so that works with both mclust version 4.4 and upper.
Other bug fixes and improvements.

Partial rewriting of the package.
“greedy” search has option for forward and backward direction.
“headlong” search has option only for forward direction in this release.
In clustvarsel() argument G is not the maximum number of clusters but it must be a vector of number of cluster to look for.
No separate code for sampling and no-sampling version of each search algorithm.
Inclusion of argument hcModel to control the initial hierarchical clustering.
Include subset selection in the regression of proposed variable on the variables already included.
“greedy” search algorithms can be executed either sequentially or using the parallel computing facilities available in R.
This version of the package requires R (>= 3.0.0) and mclust (>= 4.0).