I will introduce methods to pgmpy to select Bayesian models based on data sets. First, I will implement support for basic score-based and constraint-based structure learning. Second, I will add common enhancements to the score-based approach, including local score computation + memoization and tabu lists. Finally, I will implement the MMHC algorithm, which combines the score-based and the constraint-based method.
I believe that structure learning is the missing feature in pgmpy at the moment.