Probabilistic reasoning in intelligent systems: networks of plausible inference
Probabilistic reasoning in intelligent systems: networks of plausible inference
Theory refinement on Bayesian networks
Proceedings of the seventh conference (1991) on Uncertainty in artificial intelligence
| Hi-index | 0.00 |
The marginal likelihood of the data computed using Bayesian score metrics is at the core of score+search methods when learning Bayesian networks from data. However, common formulations of those Bayesian score metrics depend of free parameters which are hard to asses. Recent theoretical and experimental works have also shown as the commonly employed BDeu score metric is strongly biased by the particular assignments of its free parameter known as the equivalent sample size and, also, as an optimal selection of this parameter depends of the underlying distribution. This sensibility causes that wrong choices of this parameter lead to inferred models which do not properly represent the distribution generating the data even with large sample sizes. To overcome this issue we introduce here an approach which tries to marginalize this free parameter with a simple averaging method. As experimentally shown, this approach robustly performs as well as an optimum selection of this parameter while it prevents from the choice of wrong settings for this widely applied Bayesian score metric.