Abstract
The paper studies empirically the time-space trade-off between sampling and inference in the cutset sampling algorithm. The algorithm samples over a subset of nodes in a Bayesian network and applies exact inference over the rest. As the size of the sampling space decreases, requiring less samples for convergence, the time for generating each single sample increases. Algorithm wcutset sampling selects a sampling set such that the induced-width of the network when the sampling set is observed is bounded by w, thus requiring inference whose complexity is exponentially bounded by w. In this paper, we investigate the performance of w-cutset sampling as a function of w. Our experiments over a range of randomly generated and real benchmarks, demonstrate the power of the cutset sampling idea and in particular show that an optimal balance between inference and sampling benefits substantially from restricting the cutset size, even at the cost of more complex inference.