Segment 27. Mixture Models

From Computational Statistics (CSE383M and CS395T)
Jump to navigation Jump to search

Watch this segment

(Don't worry, what you see statically below is not the beginning of the segment. Press the play button to start at the beginning.)

{{#widget:Iframe |url=http://www.youtube.com/v/9pWnZcpYh44&hd=1 |width=800 |height=625 |border=0 }}

The direct YouTube link is http://youtu.be/9pWnZcpYh44

Links to the slides: PDF file or PowerPoint file

Problems

To Calculate

The file Media:Mixturevals.txt contains 1000 values, each drawn either with probability <math>c</math> from the distribution <math>\text{Exponential}(\beta)</math> (for some constant <math>\beta</math>), or otherwise (with probability <math>1-c</math>) from the distribution <math>p(x) = (2/\pi)/(1+x^2),\; x>0</math>.

1. Write down an expression for the probability of the file's data given some values for the parameters <math>\beta</math> and <math>c</math>.

2. Calculate numerically the maximum likelihood values of <math>\beta</math> and <math>c</math>.

3. Estimate numerically the Bayes posterior distribution of <math>\beta</math>, marginalizing over <math>c</math> as a nuisance parameter. (You'll of course have to make some assumption about priors.)

To Think About

1. In problem 3, above, you assumed some definite prior for <math>c</math>. What if <math>c</math> is itself drawn (just once for the whole data set) from a distribution <math>\text{Beta}(\mu,\nu)</math>, with unknown hyperparameters <math>\mu,\nu</math>. How would you now estimate the Bayes posterior distribution of <math>\beta</math>, marginalizing over everything else?