Crucial components in probabilistic inference pipelines

In state-of-the-art probabilistic logic and learning frameworks, such as ProbLog, inference is reduced to efficient Weighted Model Counting. To do so ProbLog employs a set of consecutive transformation steps, called an inference pipeline. Each step in the probabilistic inference pipeline can be conducted by different tools or algorithms, i.e. components. The choice of the components can be crucial to the performance of the system. In this paper we describe in details different ProbLog pipelines and investigate which are the crucial components with respect to the performance. Our main contributions are the thorough analysis of the inference pipelines and the introduction of new pipelines, one of which performs very well on our benchmarks.status: publishe