I think K is usually interpreted as the # of gaussians from which the data has been assumed to be sampled. Not immediately related to # of latent variables unle

I think K is usually interpreted as the # of gaussians from which the data has been assumed to be sampled. Not immediately related to # of latent variables unless you invoke kernel kmeans or something like laplacian eigenmaps/diffusion maps.