<< Click to Display Table of Contents >> Navigation:  Building blocks of QGeNIe > Inference algorithms > Introduction to QGeNIe inference algorithms

The default QGeNIe algorithm for model updating, the clustering algorithm is exact, very efficient, and fast.

Because probabilistic reasoning is worst-case NP-hard, a QGeNIe user may encounter networks for which the memory requirements or the updating time may be not acceptable. In such a case, it may be necessary to choose a different algorithm. Even though we have never experienced the need in practice to change the default algorithm, it is conceivable that the clustering algorithm will run out of memory or will take too much time. In such a case, we recommend trying the Relevance-based decomposition algorithm, which is also an exact algorithm making use of the structural properties of the model at hand and with some overhead can make the model tractable. If this does not work, we recommend the EPIS sampling algorithm, which is quite likely the most efficient stochastic sampling algorithm for discrete Bayesian networks. While it is an approximate algorithm, it shows excellent convergence rates and should deliver precision that will be satisfactory. Tradeoff between precision and computation time can be controlled by selecting the number of samples.