Nov 12
I'd like to spend one more day on the hidden markov models,
hopefully finishing this stuff up so we can move on.
The topics that I would like you to have under your belts are
- the forward algorithm, and what it's for
- the Viterbi algorithm (a small variation), and what it's for
- the particle filter approach (a numerical approximation; analogous to the Gibbs Sampling we did before)
- four types of chocolate: plain, almond, coconut, both
- two levers in control room: (almond,coconut)
- monkey in control room has 30% chance of flipping a lever
- observed color of chocolate is light or dark, P(light|plain,almond,coconut,both) is (0.1,0.3,0.8,0.9)
- a priori lever probabilities are equal
then
- a) The probability of getting (light, dark, light, dark) is what?
- b) If you see (dark, dark, dark), what is the mostly likely sequence of candies?
- c) If you see (dark, light), what is the probability that i) the first is almond and ii) the second is coconut ?
We discussed this in class, and worked out the
numbers within an ipython notbook, which I have attached.
The particle filter approach is not really all the appropriate for discrete
tiny problems like this one. BUT it can be implemented, as an illustrative example.
So ... let's describe how to do so, and let you code that for part of next week's homework.
![[paper clip]](/courses/source/wiki_images/paper_clip_tilt.png)
name last modified size
