So I recently started learning about Probabilistic Models and it is pretty confusing to me.
My understanding is that ancestral sampling makes one pass through a directed pass, conditioning a sample of one variable on previous variables, while Gibbs' Sampling uses Markov chains and multiple passes. The thing I am confused about is how you can actually sample from an energy-based model. How exactly does conditioning one variable on other variables in the graph help to sample from it? For example when you have P(y | x), I understand that x and y are neighbors in the graph, but what actually happens? How is x drawn, and how does this help compute y in the graph?
