As structure learning algorithms output either a pgmpy.base.PDAG or pgmpy.base.DAG object, to fit parameters to these models users currently need to convert them to a pgmpy.models.BayesianNetwork object. By moving the BayesianNetwork.fit method to DAG.fit, it would make the workflow simpler and more straightforward. As BayesianNetwork inherits DAG, there would be no change in the Bayesian Network API.