BENCHMARKING OF ROOT BOOSTED DECISION TREES AND XGBOOST

ROOT is a data analysis framework used in many different experiments across CERN, and beyond. The TMVA library, which forms part of ROOT, is in particu- lar a machine learning toolkit for multivariate analysis. In particular, amongst its various machine learning tools, it provides a robust boosted decisions trees (BDT) implementation. The primary task at hand was then to benchmark the CPU and memory performance of TMVA's BDT implementation, and see how it fairs. Ideally, we would benchmark against some known implementation which is con- sidered to be the 'gold standard', which in this case is XGBoost. Different machine learning libraries have different specifications on how data is to be formatted for input. To this extent, besides benchmarking, we also required a means by which we can convert training and testing data–sets prepared by TMVA into a format which is readable by XGBoost. Lastly, we required that any written benchmarks should employ the use of Google Benchmark and be integrated into rootbench, a benchmarking library for ROOT.