Divide the current ML benchmark into three major steps: 1. E/p cut study, 2. ML data preprocessing (including E/p cut), 3. ML training and benchmarking.
Improve how the python scripts handle the data to avoid insane usage of memories with big data.
Implement a 2d scan of E/p cut so it automatically chooses the best cut on [layer, E/p].
Add a script for ML data preprocessing so the benchmark can directly take standard reconstruction output (no need for MLSorter/MLCombiner which are not used in other reconstruction option files).