HPC project PoP's proof of concept for BPMF mathematical code leads to around 40% runtime reduction
14 Jun 2017 - The Bayesian Probabilisitc Matrix Factorization (BPMF) is an efficient method to solve these kind of complex modelling problems. The public available BPMF code was analysed in a POP Audit and Performance Plan service activity. While the BPMF code that showed already good scalability over several 100 nodes and also good efficiency on the node level, POP experts could still identify points for improvement.
The POP Proof of Concept changes for BPMF were evaluated with three different datasets achieving speed-ups between 1,6 and 1,8 that correspond to runtime reductions between 38 and 44%.