On the Practical Significance of Hypertree vs. TreeWidth
Rina Dechter, Lars Otten, Marinescu Radu · Frontiers in artificial intelligence and applications · 2008
The recently introduced notion of hypertree width has been shown to provide a broader characterization of tractable constraint and probabilistic networks than the tree width. This paper demonstrates empirically that in practice the bounding power of the tree width is still superior to the hypertree width for many benchmark instances of both probabilistic and deterministic networks.