annawoodard / NPFit

Effective field theory + ttV

This is the code used for the Effective Field Theory interpretation in the CMS TOP-17-005 analysis (paper submitted to JHEP.) The preprint is available here. Code used for running Madgraph to calculate cross sections is located in a separate repository which can be found here.

Installation

CCTools is required. At Notre Dame, add this to your login script:

cctools=lobster-148-c1a7ecbd-cvmfs-0941e442
export PYTHONPATH=$PYTHONPATH:/afs/crc.nd.edu/group/ccl/software/x86_64/redhat6/cctools/$cctools/lib/python2.6/site-packages
export PATH=/afs/crc.nd.edu/group/ccl/software/x86_64/redhat6/cctools/$cctools/bin:$PATH

If you are not running at Notre Dame, you will need to setup cctools yourself:

cd $HOME
slc=$(egrep "Red Hat Enterprise|Scientific|CentOS" /etc/redhat-release | sed 's/.*[rR]elease \([0-9]*\).*/\1/')
wget http://ccl.cse.nd.edu/software/files/cctools-lobster-148-c1a7ecbd-cvmfs-0941e442-x86_64-redhat$slc.tar.gz
tar xvf cctools-lobster-148-c1a7ecbd-cvmfs-0941e442-x86_64-redhat$slc.tar.gz
# add following line to your login script
export PATH=$HOME/cctools-lobster-148-c1a7ecbd-cvmfs-0941e442-x86_64-redhat$slc/bin:$PATH

Now set up a working area:

curl https://raw.githubusercontent.com/annawoodard/NPFit/master/setup.sh; source setup.sh

Quick start

Example cross section scans are included in the data directory. To reproduce the TOP-17-005 plots, only one modification of the configuration file is necessary (with caveats [1].) Edit indir in data/config.py to point to data/cross_sections/13TeV/1d.npz. After cmsenving in your working area, setup the output directory and produce the Makeflow specification:

run make data/config.py

Follow the instructions which will be printed to the screen. They should look similar to this:

# to run, issue the following commands:
cd /afs/crc.nd.edu/user/a/awoodard/www/ttV/1
nohup work_queue_factory -T condor -M ttV_FTW -C /afs/crc.nd.edu/user/a/awoodard/releases/CMSSW_8_1_0/src/NPFit/NPFit/data/factory.json >& factory.log &
makeflow -T wq -M ttV_FTW --shared-fs '/afs'

Note that you only need to run the work_queue_factory command once; you can leave it running and it will only submit jobs as they are needed for any makeflow process matching label in test/config.py (there is no reason to change the label between runs).

[1] Caveats:

• If you have a www directory which is not located at $HOME/www, you should modify outdir in the config. A web-enabled output directory is not a requirement but makes viewing plots more convenient.
• If you are not using the HTCondor batch system, modify batch type (supported systems: local, wq, condor, sge, torque, moab, slurm, chirp, amazon).
• If you are submitting to machines which will not have your output directory mounted, change outdir shared to False.

More details

Main idea

Reproducibility

Calling run make data/config.py produces an output directory (specified as outdir in the config) which contains:

1. all of the inputs (starting from datacards and including the config itself),
2. all of the outputs (plots, etc),
3. all of the commands run (specified in Makefile), and
4. a git patch in order to exactly reproduce the code as it was the last time makeflow was called. Instructions for reproducing the code are saved to outdir in a text file called README.txt, which should contain something like the following:

   cd /afs/crc.nd.edu/user/a/awoodard/releases/CMSSW_8_1_0/python/NPFit/NPFit
   git checkout d7bb6f7
   git apply /afs/crc.nd.edu/user/a/awoodard/www/ttV/1/patch.diff
   
   

It is recommended that you iterate the version number at the end of outdir periodically. In this way, you should be able to keep track of differences as the analysis evolves. Note that makeflow works similarly to a makefile; if you delete an output and run makeflow again, all of the outputs which depend on that file will also be re-run.