Does Infer.NET support Hidden Markov Models (HMMs)?
Infer.NET 2.4 beta does support Hidden Markov Models, except that they need to be ‘unrolled’ i.e. you cannot have random variable arrays for the variables along the HMM, they need to be separate variables.This unrolling means that compiling very large HMMs will be costly in both memory and time, and that also the resultant inference will be slightly slower. The click through model is a customised form of HMM, which is implemented as an unrolled set of variables.In this case, the number of variables is small and so the unrolling is not a problem. In future, we will be looking into support for rolled-up HMM models, which will make HMMs and similar chain-like models more efficient and allow them to be used for larger data sets. Inference • What inference algorithms does Infer.NET support? Infer.NET supports expectation propagation (including belief propagation as a special case), variational message passing (also known as variational Bayes, variational EM) and block Gibbs sampling. Suppor
