ReactiveMP.jl
ReactiveMP.jl copied to clipboard
Implement Gaussian Mixture Model distribution
For some applications, we might be interested in using mixtures of Gaussians for modeling data. Therefore this feature would be a nice addition to the toolbox.
Doesn't this already exist? The NormalMixture
node? Or do you mean extracting the pdf?
For a project that I am working on, I would like to be able to send out mixture distributions as messages. Although this PR mentions Gaussian mixture messages only, it might generalize to a mixture distribution of arbitrary distributions. The NormalMixture
node currently only supports a variational approximations, because it's output message is currently constrained to be a Gaussian distribution. However, for exact inference, this outgoing message would ideally be a mixture distribution.