This post looks at finite mixtures of normals. The first bit of code here looks at the flexibility of mixtures of normals. Feel free to change the argument to the plots function to try more mixtures.
We now look at estimating the model using JAGS. This first versions shows identifiability issues that arise due to label switching.
The non-identifiability can be seen by the observations having the same probability of coming from each component as well as the near identical posteriors for mu and sigma from each component.
In order to make the parameters identified, we order the means using the
sort function in JAGS.
blog comments powered by