EM algorithm details

Johan · Post by **Johan** » Thu Jul 22, 2010 5:26 pm

Hi,

I just want to obtain some details regarding the EM algorithm in SMILearn.
1. What does the equivalent sample size set? Does this algorithm use Dirichlet distribution as priors (which has the concept of equivalent sample size) as I know a popular algorithm does or is it something different?
2. what does randomising the parameters do?

mark · Post by **mark** » Wed Aug 04, 2010 6:06 am

1. Your guess is correct.
2. It randomizes all the CPT parameters (but only if the equivalent sample size is zero).