Reproducing the 2019 joint analysis » History » Version 9

« Previous - Version 9/18 (diff) - Next » - Current version
Ashley Back, 02/04/2019 04:04 PM
Adds details on merging job output files.

Reproducing the 2019 joint analysis

Making new predictions

The framework for generating analysis predictions with the systematic shifts is undergoing a bit of an overhaul, but its usable in many cases already. The current workflow is based on the code committed in CAFAna/shared/syst_framework.
The main interface is through a python wrapper script which parses various input options such as the analysis, beam configuration, list of systematics etc. For more info on the options : -h 
usage: [-h] -a ANALYSIS -b BEAM [-c] [-nt] [-fk]
                      [-sg SYST_GROUP [SYST_GROUP ...]] [-all]
                      [-xp EXTRAPS [EXTRAPS ...]] [-all_xp] [-g [GRID]]
                      [-i [INTERACTIVE]]

Submit jobs for generating analysis predictions

optional arguments:
  -h, --help            show this help message and exit
  -a ANALYSIS, --analysis ANALYSIS
                        Specify analysis, numu2018/nue2018/nus2018
  -b BEAM, --beam BEAM  Specify beam mode, FHC/RHC
  -c, --concat          Specify whether to use concats
  -nt, --notau          Do not use tau swap files
  -fk, --fake_data      Use fake data at ND
  -sg SYST_GROUP [SYST_GROUP ...], --syst_group SYST_GROUP [SYST_GROUP ...]
                        Specify syst groups to generate
  -all, --all_systs     Generate predictions for all syst groups
  -xp EXTRAPS [EXTRAPS ...], --extraps EXTRAPS [EXTRAPS ...]
                        Specify extrapolations to carry out
  -all_xp, --all_extraps
                        Generate predictions for all extrapolations
  -g [GRID], --grid [GRID]
                        Run this on the grid with options
  -i [INTERACTIVE], --interactive [INTERACTIVE]
                        Run this interactively with cafe options

For example, one can generate predictions interactively over a single concat file, for the LightLevelUp systematic in FHC mode for the nue2018 analysis by simply running : -a "nue2018" -b "FHC" -c -sg 'LightUp' -all_xp -i='-bq -l1 -ss'

The python wrapper picks up the relevant analysis components from syst_header.h in the same folder and writes a CAFAna macro make_predictions_systs.C on the fly based on a template macro. This is similar in behaviour to make_sim_fcl.
Right now, it only supports interactive running but the grid option will be implemented soon.

If you want to generate systematically shifted predictions on the grid right now, you can use the old framework in CAFAna/shared/Ana2019 that contains a static copy of make_predictions_systs.C whose behaviour you control by passing text string options to the macro. For example, generate the same predictions as we did, using above, you would run:

cafe -bq -l 1 -ss make_predictions_systs.C FHCLightLevelUp_nueconcat_realNDData allxp_nue

To run this on the grid you would then just pass it to You can find a more detailed example of how to run this in the accompanying PDF for the tutorial (docdb-35464).

If you are generating systematically shifted predictions on the grid, you should end up with a number of files per systematics group for which you generate predictions. You should have the same number of files as jobs submitted multiplied by the number of extrapolations you ran (since each decomp/extrap option generates a separate prediction root file). We have a useful build script to ensure you combine job output root files and root files for different decomp/extrap options in the correct order. The eventual goal is for to autogenerate an accompanying build script to merge all the prediction files it outputs. But until this mechanism works correctly you can use in CAFAna/shared/syst_framework, and just edit the directory string at the top of the script. Running the script is then as simple as:


All merged files are output to a merged_predictions directory. To clean or rebuild the files in merged_predictions, run ./ clean or ./ rebuild. Again you can find a more detailed explanation of this tool in (docdb-35464).

Performing the joint fit

Making supporting plots

Additional material/notes

Editing systematics code

Rudimentary examples with more explanation can be found in the tutorial macro source:/trunk/CAFAna/tute/demoSysts.C

Videos from the 2019-Analysis workshop:

  • Liudmila Kolupaeva's tutorial on how to do the joint fit: (see file zoom_0.mp4 below)
  • Andrew Sutton's tutorial on how to do the joint fit: (see file zoom_1.mp4 below)