[Insight-users] Expectation maximization Algorithm-Stopping Rule?

[Ehtiati, Tina (SCR US EXT)]

Hello,
 
I have some questions related to the
ExpectationMaximizationMixtureModelEstimator class.
 
 I have been running the example code provided for this algorithm and
according to the example the 
standard deviation for the two Guassian components are chosen as (both)
30 when generating the sample data points. The initializiation values
for the standard deviations for the EM algorithm are 800, and 850.
After running the example program (number of iterations=200) the results
are as the following :
 
Cluster [0]
Parameters: [101.886 , 782.963]
Proportion:0.5
 
Cluster[1]
Parameters:[198.538, 951.226]
Proportion: 0.49
 
Does that mean that the standard deviation has not been estimated
correctly and the program has to run more iterations to converge to
correct answers for the standard deviations? And this brings me to my
main question:
 
Is there a built-in way to choose another rule for stopping the EM
algorithm, other than giving it a fixed number  of iterations, for
example setting a threshold for result error rates or some other
stopping rules? Or shall I call the estimator inside a loop and keep
checking the results untill satisfactory?  
 
I appreciate your help and any explanation.
 
Thanks a lot
 
Tina
 
 
 
 
 
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20070620/b420a575/attachment.htm