Summarizing and visualizing a set of bayesian networks with quasi essential graphs

Overview & Synopsis

Many learning methods now generate a set of models in order to improve robustness. Evaluating for instance the quality of a set of Bayesian networks is quite usual for estimating separately the quality of each model and for summarizing these results. Visualizing the outcomes are a more complex task. We propose in this work an approach based on an inverse principle. Firstly, we build the Quasi Essential Graph (QEG), "most" representative of the whole set. Then, we apply the usual quality operators for this new object. This paper describes the notion and properties of Quasi Essential Graph. An algorithm for its extraction is proposed, as well as a graphical metaphor for its visualization. A toy example is finally given for the sake of illustration.