CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: Benchpress: A scalable, platform-independent workflow to benchmark structure learning algorithms for graphical models Authors:  Felix Rios - University of Basel (Switzerland) [presenting]
Giusi Moffa - University of Basel (Switzerland)
Jack Kuipers - ETH Zurich (Switzerland)
Abstract: Probabilistic graphical models (PGMs) play a central role in statistical data analysis, thanks to their compact and elegant way to represent complex dependence structures in multivariate probability distributions. In many realistic situations, ranging from disciplines such as social sciences to epidemiology, medicine, and biology, researchers are interested in finding the structure of an underlying model. Structure learning of PGMs is a computationally intensive task, with many and varied algorithms in constant development. The fast-moving pace and the heterogeneity of state-of-the-art algorithms pose a practical challenge for many researchers who wish to choose the most suitable algorithm for their specific problem or compare the performance of a novel method to the current state of the art. We will present a novel snakemake workflow, benchpress, which provides a unified platform for reproducible and scalable benchmarking and execution of structure learning algorithms for PGMs. Benchpress provides a platform where researchers and practitioners can easily compare existing algorithms available in the public domain in their original implementation, as well as enable other researchers to reproduce their results easily. We will demonstrate some of the functionalities in several analysis scenarios, typical for researchers and data scientists. The source code and documentation are publicly available from