EM algorithm : how to have prior?

Question

I want to use EM algorithm for Gaussian Mixture Model estimation.

I know my data comes from 2 sources A and B.

To perform EM algorithm, I need to input the prior probabilities P(A) and P(B).

My question is : should I use P(A) = P(B) = 0.5 ? Or should I use prior estimation of P(A) and P(B) ?

Answer 1

For EM, you can either initialise each datapoint to one of the clusters using random initializations or K means. From this initialization, we use the sample means as estimates of corresponding Gaussian mean parameter, similarly for the covariance matrix, to start the first iteration. This is a frequentist approach.

Priors are not required for a vanilla EM.