https://docs.pymc.io/notebooks/gaussian-mixture-model-advi.html
Could you elaborate “class assignment variables are marginalized out, giving weighted sum of the probability for the gaussian components”?
What it meant there is that we do not model the latent label with a discrete variable, but a mixture model. There are a bit more elaborated text from the Stan user manual: https://mc-stan.org/docs/2_21/stan-users-guide/latent-discrete-chapter.html