bayesian network
1. conditioned on evidence
- Can do rejection sampling – only accept samples where the evidence \(e\) is true.
- Why can't you just set force the evidence node to be true? Because then you're sampling procedure won't be consistent. If we have nodes \(A\), \(B\), and edge \(A \rightarrow B\), then forcing \(B\) to be true and sampling from the network is going to result in exactly the same distribution over \(A\). See these slides.