Christos Tzamos
Email:
@mit.edu
tzamos
Phone:
9091739
(617)
Office:
32G630, Stata Center
Web:
christos.me
Google Scholar 
DBLP 
arXiv 
Github

The following graph shows how the EM the steps of the EM algorithm for a NaiveBayes setting in a population model. In the example, there are n_{1}+n_{2} features each of which is an independent Bernouli variable. There are two document classes C1 and C2 occuring with equal probability. C1 takes value 1 in each of the first n_{1} features with probability p_{1} and takes value 1 in each of the n_{2} features with probability p_{2}. C2 has complementary probabilities in each feature, that is, if C1 has probability p on a feature C2 has probability 1p.
The graph shows the two stable fixed points corresponding to the probabilities of the classes. It also draws a separator for their regions of convergence. Clicking anywhere of the graph runs the EM algorithm for 100 steps with guesses for (p_{1},p_{2}) begining from (x,y).