# EM Algorithm for Bayesian Networks with missing data

Setting: learning parameters of Bayesian Network (BN) with missing data.

Algorithm: Expectation-Maximization.

Question: suppose I am in the M-step, and that in the complete data there are no examples for one class. In this case, how can I update the conditional probability for that specific class?

I understand that the question is quite vague. Thus, please see slide 15: how can I get $$p(X_3=2 | X_2=1)$$ if there are no examples in the complete data?

• The complete data is integrated over the latent variable, hence over all the possible values of this variable, conditional on the observed data. – Xi'an Jan 2 at 16:48