User Guide# Columnflow Structure General introduction Tasks in columnflow Important note on required parameters Important modules and configs Analysis, Campaign and Config Building Blocks Calibrations Selections Introduction Create an instance of the Selector class Running the SelectEvents task Production of columns Introduction Usage ProduceColumns task Categories Create a category Creation of nested categories Extend your categorization as part of a Producer Helper functions for categories Key points (TL; DR) Config Objects Generalities Parameters from the order package (required) Auxiliary Parameters (optional) Sandboxes and Their Use in Columnflow What needs to be defined Where to Store the Corresponding Files The Setup File The requirement file How to Use the Namespace Machine Learning How training happens in Columnflow: K-fold cross validation Configure your custom machine learning class: After the configuration ABC functions sandbox: datasets: produces: uses: output: open_model: train: evaluate: Commands to start training and evaluation Optional useful functions Separate training and evaluation configuration for configs, calibrators, selector and producers setup Adjusting the ML model settings via the command line Systematic Uncertainties Uncertainties that only modify the yield Weight-based uncertainties Uncertainties that modify the selection efficiency Uncertainties with dedicated datasets Plotting Creating your first plot Customization of plots Creating 2D plots Creating cutflow plots Creating plots for different shifts Directly displaying plots in the terminal Using your own plotting function Applying a selection to a variable Creating your own task Writing new tasks Example: Task Class The “super” class The sandbox The parameters The requirements RunOnceTask Add the task to the analysis Executing the Task Custom law.cfg file for users without grid certificate (CMS specific) Debugging Debugging outputs Debugging outputs of supported extensions (ROOT, Parquet, JSON and Pickle) Debugging histograms FAQ Troubleshooting: Law introduction Tasks and parameters Law Config Running remote Best practices Selecting output locations Controlling versions of upstream tasks Task family specific parameters Pinned versions in the analysis config or law.cfg file Dynamic variation of used columns using DeferredColumn Columnflow convenience tools General advices Using python scripts removed from the standard workflow Special Usecases Mechanism to hook into histograms (e.g. for ABCD methods) Hist hooks Setting default Calibrator, Selector, Producer, Ml-Model (CSPM) via function CMS specializations Tasks CreatePileUpWeights
