Hyperparameter sensitivity revisited
[edit]
Proceedings of The 3rd International Workshop on Advanced Methodologies for Bayesian Networks, PMLR 73:77, 2017.
Abstract
The BDeu scoring criterion for learning Bayesian network structures is known to be very
sensitive to the equivalent sample size hyperparameter. Recently some authors have suggested
alternative Bayesian scoring criteria that appear to behave better than BDeu. So
is the problem solved? We will review the problem and suggested solutions and present
empirical assessment of the current situation.
Related Material


