I found a problem with the two metrics(average test set log-likelihood for each subtype model and CVIC) and CVIC) for selecting the best number of subtypes. I have found some differences in the way they calculate, and sometimes the results of using them to select the best number of subtypes are different. At this time, which metrics is more reasonable.
I found a problem with the two metrics(average test set log-likelihood for each subtype model and CVIC) and CVIC) for selecting the best number of subtypes. I have found some differences in the way they calculate, and sometimes the results of using them to select the best number of subtypes are different. At this time, which metrics is more reasonable.