Metropolis-Hastings inference of local variables


#1

Hello,

When applying MH to a model with both local and global latent variables, the acceptance ratio is always 0. I saw somewhere and agree that since a sample for the local variables is the size of the number of samples, a proposal for these will almost certainly be rejected. But if I integrate out the latent variables and perform inference only over the global ones, the acceptance is still 0! Why?