Is it possible to learn the structure of a Bayesian network using PyMC? I have ~100 examples and all features are discrete. Here is an example of causal modeling, but the structure is already known. In my case, I know the priors of just one variable, the root.
I think the short answer is no. Pymc is not really designed for Bayesian networks but for Bayesian data analysis, which are not the same thing. There is some overlap in that both use a DAG structure for inference, but there are no structure learning algorithms like PC available in the API. With continuous random variables, you cannot simply switch the inferential direction and easily calculate the likelihood of the data.