CMS dijet anomaly detection dataset

This record contains simulated dijet events released by the CMS Collaboration to support the development of anomaly detection techniques. It consists in an unweighted sample of standard model background events and matching signal events. The variables, which describe the substructure of the two leading jets, include high-level observables such as subjetiness ratios as well as the four-momenta of up to 100 jets constituents. Inputs needed to compute uncertainties are also included for signal events. The detector simulation correspond to years 2016–2018. The simulation, preprocessing, and file format are described in the related publication. The files background_0.h5 through background_19.h5 contain 24 million background events, which corresponds to an effective luminosity of about 27/fb. Other h5 files each contain events generated for a signal model considered in the paper. They follow the naming convention of the original datasets, a subset of which can be found on the CERN Open Data Portal together with the production cards. The name of the datasets used to construct the background sample are listed in the file background_samples.txt. All source datasets openly available at the time of publication are also linked at the bottom of this page. Starting from the MiniAOD format, the events have first been converted to the PFNano format. Commit bbb8657c2e223faf38557d3caca304716da8c813 was used in CMSSW_10_6_X (CMSSW usage instructions can be found in the CMS Open Data Guide). The preselection and conversion to h5 format are performed by running the code provided in H5_maker.zip in the same CMSSW release; usage instructions can be found in the corresponding README.