Speakers
Description
We present a prototype implementation of a particle physics analysis workflow using Snakemake for an ATLAS anomaly detection search. Snakemake provides a flexible and scalable workflow for managing thousands of jobs with complex dependencies, supporting execution both locally and across different HPC environments. The workflow cleanly separates small-scale tasks, such as plotting, histogram manipulation, or small data transformation from large-scale computations including neural network training by running these tasks interactively or submitting them as HPC jobs, as appropriate. Snakemake configurations allow running on different HPCs with minimal additional effort, enabling collaboration while utilizing local cluster resources. Snakemake is currently used in ATLAS to orchestrate an anomaly detection search. Snakemake enables systematic scans over different random seeds, mass regions, signal models and injected signal strengths, all of which are crucial ingredients of such an analysis. By automating the entire analysis chain, from the initial ROOT ntuples to the final results, the workflow improves reproducibility, transparency and onboarding, providing a robust and maintainable structure for modern data-intensive analyses.