Initialization in TidyLPA

26 views

Skip to first unread message

Peruna FF

unread,

Sep 22, 2021, 9:58:07 AM9/22/21

to tidyLPA

Thanks for TidyLPA developer. It has helped a great deal for my current project.

I have run my classification model using TidyLPA and via mclust directly. I noticed that the default set-up of the mclust (i.e., Mclust(dat)) is very sensitive to the initial values but fast, whereas TidyLPA with same model specification gives much stable output but takes longer time.

I understand that mclust only use a single set of start value, and the risk of local optimum is not addressed (see a reference here) . What is TidyLPA strategy to avoid local maxima? Given the model is much slower to run, I guess additional steps may have been added from the basic mclust.

I will much appreciate it if any of you could give me a lead. Many thanks in advance for your help.