More Gibbs Sampling

These are lecture notes from my Computer Science course.

Assessment info

The assessment will not cover anything from now on; will cover from lecture 14 (?) back. No lecture Tuesday Week 10

 More/recap on Gibbs sampling

Gibbs sampling with already instantiated variables doesn’t always work. For example: take a BN A->B where A is a uniform distribution (50/50) and B is the value of A with probability 1. If you start with A=0;B=0, the Markov chain will be stuck there even though P(A=0.5). This is known as non-ergodic.

What about continuous sampling?

You can use it for this. You can use it where you have a mix too.

This entry was posted in agm, lecture. Bookmark the permalink.