Hey, I implemented a gaussian mixture model over the Iris dataset (with 3 different features / clusters).
I am not sure how to deal with “label switching” issue in my model.
I have seen several techniques to deal with this issue, among them is constraints over the prior, i.e to formulate the prior in such a way as to ensure there is only one posterior mode (eg. order the means of the mixture components), but I understand that this technique isn’t generally used.
Another strategy is to ignore the problem during sampling, and then post-process the output to re-label the components to keep the labels consistent , which I am not sure how to implement.
I couldn’t find any solution in the NumPyro documentation or anywhere else.
I would really be glad to get some help with this issue.