Class 10: More Simulation and the EM Algorithm
- Debugging tools
-
You'll get better with them the more you use them.
- EM algorithm
-
When dealing with exponential families, the algorithm amounts to
- Estimate "complete" sufficient statistics given observations
and current parameter estimates.
- Maximize the associated likelihood to obtain new parameter
estimates, and iterate.
Today, we see it in a more general application where the usual
sufficiency does not apply.
Status of Projects
- I will check your pages tomorrow for their current status. You should have
done some sort of Mathematica calculation for your pages and incorporated it into
the page.
- My page is slowly improving, with
more links to other sites.
- Smoothing
- We will start with an example relevant to many of the projects that
you have chosen, namely smoothing in regression.
- Mixtures
- My favorite use of the EM, but watch out for identification problems. Here
we'll do some examples of normal mixtures.
- Lisp script for today's class
- class10.lsp
Next time
Hopefully, I will finally be ready to say something about Gibbs samplers.